Search

Next-generation single-molecule protein sequencing technologies have the potential to significantly accelerate biomedical research. These technologies offer sensitivity and scalability for proteomic analysis. One auspicious method is fluorosequencing, which involves: cutting naturalized proteins into peptides, attaching fluorophores to specific amino acids, and observing variations in light intensity as one amino acid is removed at a time. The original peptide is classified from the sequence of light-intensity reads, and proteins can subsequently be recognized with this information. The amino acid step removal is achieved by attaching the peptides to a wall on the C-terminal and using a process called Edman Degradation to remove an amino acid from the N-Terminal. Even though a framework (Whatprot) has been proposed for the peptide classification task, processing times remain restrictive due to the massively parallel data acquisicion system. In this paper, we propose a new beam search decoder with a novel state formulation that obtains considerably lower processing times at the expense of only a slight accuracy drop compared to Whatprot. Furthermore, we explore how our novel state formulation may lead to even faster decoders in the future. Author summary: Proteomic analyses frequently rely on mass spectrometry, a method characterized by its limited dynamic range, potentially overlooking low-abundant proteins. To address this limitation, single-molecule protein sequencing methods offer a solution. Fluorosequencing is a cutting-edge single-molecule protein sequencing method, which can distinguish peptides or protein molecules massively parallelly. This method has attracted interest from investors, as evidenced by the recent funding of Erisyon, a company developing this technology. This technique contains a challenging classification task: determining the original peptide sequence from light-intensity observations obtained after several Edman cycles. A classifier based on a combination of k Nearest Neighbors (kNN) with Hidden Markov Models (HMM) had been shown to have close-to-optimal accuracy with tractable complexity. We propose in this paper a new algorithm that reduces computation time significantly at the expense of a slight reduction in accuracy compared to state-of-the-art method. [ABSTRACT FROM AUTHOR]

In proteomics, a crucial aspect is to identify peptide sequences. De novo sequencing methods have been widely employed to identify peptide sequences, and numerous tools have been proposed over the past two decades. Recently, deep learning approaches have been introduced for de novo sequencing. Previous methods focused on encoding tandem mass spectra and predicting peptide sequences from the first amino acid onwards. However, when predicting peptides using tandem mass spectra, the peptide sequence can be predicted not only from the first amino acid but also from the last amino acid due to the coexistence of b-ion (or a- or c-ion) and y-ion (or x- or z-ion) fragments in the tandem mass spectra. Therefore, it is essential to predict peptide sequences bidirectionally. Our approach, called NovoB, utilizes a Transformer model to predict peptide sequences bidirectionally, starting with both the first and last amino acids. In comparison to Casanovo, our method achieved an improvement of the average peptide-level accuracy rate of approximately 9.8% across all species. Author summary: Understanding the characteristics of data is very important in deep learning methods. When predicting sentences, the transformer model naturally predicts from the first word. For this reason, previous methods predicted peptide sequences from the first amino acid. However, in tandem mass spectra, it is possible to predict peptide sequences bidirectionally. This method shows better results than previous approaches because it can better encode tandem mass spectra. We have demonstrated that good results can be achieved simply by understanding the characteristics of such data and using the model appropriately. We hope that this paper will help various readers improve the performance capabilities of their models. Furthermore, given that bidirectional peptide sequence prediction is crucial in de novo peptide sequence analysis, we hope that this approach will be applied to both existing and future methods utilizing deep learning techniques. [ABSTRACT FROM AUTHOR]

The ubiquitous availability of genome sequencing data explains the popularity of machine learning-based methods for the prediction of protein properties from their amino acid sequences. Over the years, while revising our own work, reading submitted manuscripts as well as published papers, we have noticed several recurring issues, which make some reported findings hard to understand and replicate. We suspect this may be due to biologists being unfamiliar with machine learning methodology, or conversely, machine learning experts may miss some of the knowledge needed to correctly apply their methods to proteins. Here, we aim to bridge this gap for developers of such methods. The most striking issues are linked to a lack of clarity: how were annotations of interest obtained; which benchmark metrics were used; how are positives and negatives defined. Others relate to a lack of rigor: If you sneak in structural information, your method is not sequence-based; if you compare your own model to "state-of-the-art," take the best methods; if you want to conclude that some method is better than another, obtain a significance estimate to support this claim. These, and other issues, we will cover in detail. These points may have seemed obvious to the authors during writing; however, they are not always clear-cut to the readers. We also expect many of these tips to hold for other machine learning-based applications in biology. Therefore, many computational biologists who develop methods in this particular subject will benefit from a concise overview of what to avoid and what to do instead. [ABSTRACT FROM AUTHOR]

Drug-target binding affinity prediction plays a key role in the early stage of drug discovery. Numerous experimental and data-driven approaches have been developed for predicting drug-target binding affinity. However, experimental methods highly rely on the limited structural-related information from drug-target pairs, domain knowledge, and time-consuming assays. On the other hand, learning-based methods have shown an acceptable prediction performance. However, most of them utilize several simple and complex types of proteins and drug compounds data, ranging from the protein sequences to the topology of a graph representation of drug compounds, employing multiple deep neural networks for encoding and feature extraction, and so, leads to the computational overheads. In this study, we propose a unified measure for protein sequence encoding, named BiComp, which provides compression-based and evolutionary-related features from the protein sequences. Specifically, we employ Normalized Compression Distance and Smith-Waterman measures for capturing complementary information from the algorithmic information theory and biological domains, respectively. We utilize the proposed measure to encode the input proteins feeding a new deep neural network-based method for drug-target binding affinity prediction, named BiComp-DTA. BiComp-DTA is evaluated utilizing four benchmark datasets for drug-target binding affinity prediction. Compared to the state-of-the-art methods, which employ complex models for protein encoding and feature extraction, BiComp-DTA provides superior efficiency in terms of accuracy, runtime, and the number of trainable parameters. The latter achievement facilitates execution of BiComp-DTA on a normal desktop computer in a fast fashion. As a comparative study, we evaluate BiComp's efficiency against its components for drug-target binding affinity prediction. The results have shown superior accuracy of BiComp due to the orthogonality and complementary nature of Smith-Waterman and Normalized Compression Distance measures for protein sequences. Such a protein sequence encoding provides efficient representation with no need for multiple sources of information, deep domain knowledge, and complex neural networks. Author summary: Drugs alter the protein conformation and functionality through binding to the target proteins. Measuring the binding tendency between drugs and proteins, known as drug-target binding affinity prediction, is an important task in developing new drug candidates and novel therapeutics in the early stage of drug discovery. Measuring the affinity strength through biological assays is expensive and time-consuming. On the other hand, considerable data-driven and computational methods are proposed to predict binding affinity prediction through employing complicated models, and multiple types of proteins and drug compounds data. In this paper, we present a deep learning-based method, named BiComp-DTA, which utilizes both biological and sequence content features extracted from the protein sequences using a new unified measure, named BiComp. Encoded protein sequences and drug's information are passed to the fully-connected and hybrid neural networks, including CNN and separable CNN layers, respectively. Simulation results using benchmark datasets indicate that BiComp-DTA provides an accurate and fast prediction for drug-target binding affinity prediction without employing multiple complicated networks, several types of proteins and drug compounds data, and external tools for input data modeling. [ABSTRACT FROM AUTHOR]

Directed laboratory evolution applies iterative rounds of mutation and selection to explore the protein fitness landscape and provides rich information regarding the underlying relationships between protein sequence, structure, and function. Laboratory evolution data consist of protein sequences sampled from evolving populations over multiple generations and this data type does not fit into established supervised and unsupervised machine learning approaches. We develop a statistical learning framework that models the evolutionary process and can infer the protein fitness landscape from multiple snapshots along an evolutionary trajectory. We apply our modeling approach to dihydrofolate reductase (DHFR) laboratory evolution data and the resulting landscape parameters capture important aspects of DHFR structure and function. We use the resulting model to understand the structure of the fitness landscape and find numerous examples of epistasis but an overall global peak that is evolutionarily accessible from most starting sequences. Finally, we use the model to perform an in silico extrapolation of the DHFR laboratory evolution trajectory and computationally design proteins from future evolutionary rounds. Author summary: Laboratory evolution has revolutionized our understanding of protein structure, function, and evolution, and has generated countless useful proteins broad applications in medicine, biocatalysis, and biotechnology. These experiments explore protein sequence space through iterative rounds of mutation and selection and can provide rich data of populations traversing the fitness landscape. In this paper, we present a statistical learning framework that models the evolutionary process and can infer the structure of the underlying protein fitness landscape from multiple snapshots along a laboratory evolution trajectory. We generate a dihydrofolate reductase (DHFR) laboratory evolution data set and apply our modeling approach to infer the landscape parameters. The estimated parameters pinpoint key residues that dictate DHFR structure and function. We use the resulting model to understand the local and global structure of the fitness landscape and to perform in silico directed evolution for protein engineering. [ABSTRACT FROM AUTHOR]

In response to the imperfections of current sequence alignment methods, originated from the inherent serialism within their corresponding electrical systems, a few optical approaches for biological data comparison have been proposed recently. However, due to their low performance, raised from their inefficient coding scheme, this paper presents a novel all-optical high-throughput method for aligning DNA, RNA, and protein sequences, named HELIOS. The HELIOS method employs highly sophisticated operations to locate character matches, single or multiple mutations, and single or multiple indels within various biological sequences. On the other hand, the HELIOS optical architecture exploits high-speed processing and operational parallelism in optics, by adopting wavelength and polarization of optical beams. For evaluation, the functionality and accuracy of the HELIOS method are approved through behavioral and optical simulation studies, while its complexity and performance are estimated through analytical computation. The accuracy evaluations indicate that the HELIOS method achieves a precise pairwise alignment of two sequences, highly similar to those of Smith-Waterman, Needleman-Wunsch, BLAST, MUSCLE, ClustalW, ClustalΩ, T-Coffee, Kalign, and MAFFT. According to our performance evaluations, the HELIOS optical architecture outperforms all alternative electrical and optical algorithms in terms of processing time and memory requirement, relying on its highly sophisticated method and optical architecture. Moreover, the employed compact coding scheme highly escalates the number of input characters, and hence, it offers reduced time and space complexities, compared to the electrical and optical alternatives. It makes the HELIOS method and optical architecture highly applicable for biomedical applications. Author summary: The character-by-character alignment of two long biological sequences, i.e. DNA, RNA, and protein, is a tedious task, but essential for recognizing homologies, relationships, and variations. In this case, every alteration, including mutations (substitution), and indels (insertion or deletion) is vital and required for many biological developments like diagnosis, medicine, and vaccination. However, the applicability of current sequence alignment methods is limited, specifically in processing time and memory usage, due to their inherent serialism and imperfections of electrical systems, as well as inefficient coding schemes of optical approaches. It approximately leads to quadratic run-time and space requirements in terms of input sequence lengths, becoming an expensive and laborious process for the real-time alignment of large datasets. Hence, proposing a superior alignment method in terms of accuracy, performance, and applicability can promote biological research and developments. Here, we show that we can overcome the long-lasting and challenging problems in sequence alignment procedure by exploiting optics as a novel computing technology. In this manner, we propose a novel method and its optical architecture for alignment of DNA, RNA, and protein sequences by exploiting high-speed processing and operational parallelism in optics. As our simulation studies confirm, it provides an accurate sequence alignment with outperforming the most widely used electrical and optical alternatives in the terms of processing time and memory requirements. [ABSTRACT FROM AUTHOR]

Most methods for phylogenetic tree reconstruction are based on sequence alignments; they infer phylogenies from substitutions that may have occurred at the aligned sequence positions. Gaps in alignments are usually not employed as phylogenetic signal. In this paper, we explore an alignment-free approach that uses insertions and deletions (indels) as an additional source of information for phylogeny inference. For a set of four or more input sequences, we generate so-called quartet blocks of four putative homologous segments each. For pairs of such quartet blocks involving the same four sequences, we compare the distances between the two blocks in these sequences, to obtain hints about indels that may have happened between the blocks since the respective four sequences have evolved from their last common ancestor. A prototype implementation that we call Gap-SpaM is presented to infer phylogenetic trees from these data, using a quartet-tree approach or, alternatively, under the maximum-parsimony paradigm. This approach should not be regarded as an alternative to established methods, but rather as a complementary source of phylogenetic information. Interestingly, however, our software is able to produce phylogenetic trees from putative indels alone that are comparable to trees obtained with existing alignment-free methods. Author summary: Phylogenetic tree inference based on DNA or protein sequence comparison is a fundamental task in computational biology. Given a multiple alignment of a set of input sequences, most approaches compare aligned sequence positions to each other, to find a suitable tree, based on a model of molecular evolution. Insertions and deletions that may have happened since the input sequences evolved from their last common ancestor are ignored by most phylogeny methods. Herein, we show that insertions and deletions can provide an additional source of information for phylogeny inference, and that such information can be obtained with a simple alignment-free approach. We provide an implementation of this idea that we call Gap-SpaM. The proposed approach is complementary to existing phylogeny methods since it is based on a completely different source of information. It is, thus, not meant to be an alternative to those existing methods but rather as a possible additional source of information for tree inference. [ABSTRACT FROM AUTHOR]

Generative models of protein sequence families are an important tool in the repertoire of protein scientists and engineers alike. However, state-of-the-art generative approaches face inference, accuracy, and overfitting- related obstacles when modeling moderately sized to large proteins and/or protein families with low sequence coverage. Here, we present a simple to learn, tunable, and accurate generative model, GENERALIST: GENERAtive nonLInear tenSor-factorizaTion for protein sequences. GENERALIST accurately captures several high order summary statistics of amino acid covariation. GENERALIST also predicts conservative local optimal sequences which are likely to fold in stable 3D structure. Importantly, unlike current methods, the density of sequences in GENERALIST-modeled sequence ensembles closely resembles the corresponding natural ensembles. Finally, GENERALIST embeds protein sequences in an informative latent space. GENERALIST will be an important tool to study protein sequence variability. Author summary: Protein sequence families show tremendous sequence variation. Yet, it is thought that a large portion of the functional sequence space remains unexplored. Generative models are machine learning methods that allow us to learn what makes proteins functional using sequences of naturally occurring proteins. Here, we present a new type of generative model GENERALIST: GENERAtive nonLInear tenSor-factorizaTion for protein sequences that is accurate, easy to implement, and works with very small datasets. We believe that GENERALIST will be an important tool in the repertoire of protein scientists and engineers alike. [ABSTRACT FROM AUTHOR]

The design of vaccines against highly mutable pathogens, such as HIV and influenza, requires a detailed understanding of how the adaptive immune system responds to encountering multiple variant antigens (Ags). Here, we describe a multiscale model of B cell receptor (BCR) affinity maturation that employs actual BCR nucleotide sequences and treats BCR/Ag interactions in atomistic detail. We apply the model to simulate the maturation of a broadly neutralizing Ab (bnAb) against HIV. Starting from a germline precursor sequence of the VRC01 anti-HIV Ab, we simulate BCR evolution in response to different vaccination protocols and different Ags, which were previously designed by us. The simulation results provide qualitative guidelines for future vaccine design and reveal unique insights into bnAb evolution against the CD4 binding site of HIV. Our model makes possible direct comparisons of simulated BCR populations with results of deep sequencing data, which will be explored in future applications. Author summary: Vaccination has saved more lives than any other medical procedure. But, we do not have robust ways to develop vaccines against highly mutable pathogens. For example, there is no effective vaccine against HIV, and a universal vaccine against diverse strains of influenza is also not available. The development of immunization strategies to elicit antibodies that can neutralize diverse strains of highly mutable pathogens (so-called 'broadly neutralizing antibodies', or bnAbs) would enable the design of universal vaccines against such pathogens, as well as other viruses that may emerge in the future. In this paper, we present an agent-based model of affinity maturation–the Darwinian process by which antibodies evolve against a pathogen–that, for the first time, enables the in silico investigation of real germline nucleotide sequences of antibodies known to evolve into potent bnAbs, evolving against real amino acid sequences of HIV-based vaccine-candidate proteins. Our results provide new insights into bnAb evolution against HIV, and can be used to qualitatively guide the future design of vaccines against highly mutable pathogens. [ABSTRACT FROM AUTHOR]

A protein superfold is a type of protein fold that is observed in at least three distinct, non-homologous protein families. Structural classification studies have revealed a limited number of prevalent superfolds alongside several infrequent occurring folds, and in α/β type superfolds, the C-terminal β-strand tends to favor the edge of the β-sheet, while the N-terminal β-strand is often found in the middle. The reasons behind these observations, whether they are due to evolutionary sampling bias or physical interactions, remain unclear. This article offers a physics-based explanation for these observations, specifically for pure parallel β-sheet topologies. Our investigation is grounded in several established structural rules that are based on physical interactions. We have identified "frustration-free topologies" which are topologies that can satisfy all the rules simultaneously. In contrast, topologies that cannot are termed "frustrated topologies." Our findings reveal that frustration-free topologies represent only a fraction of all theoretically possible patterns, these topologies strongly favor positioning the C-terminal β-strand at the edge of the β-sheet and the N-terminal β-strand in the middle, and there is significant overlap between frustration-free topologies and superfolds. We also used a lattice protein model to thoroughly investigate sequence-structure relationships. Our results show that frustration-free structures are highly designable, while frustrated structures are poorly designable. These findings suggest that superfolds are highly designable due to their lack of frustration, and the preference for positioning C-terminal β-strands at the edge of the β-sheet is a direct result of frustration-free topologies. These insights not only enhance our understanding of sequence-structure relationships but also have significant implications for de novo protein design. Author summary: A protein superfold is a protein fold that appears in at least three different non-homologous protein families. Superfolds are unique in their ability to accommodate multiple functions within a single fold, a feature not typically seen in other folds. Studies in structural classification have led to two notable observations: the existence of a limited number of common superfolds contrasted with a larger variety of less frequent folds, and a recurring pattern in α/β type superfolds where the C-terminal β-strand often occupies the edge of the β-sheet, while the N-terminal β-strand is usually found in the middle. The origins of these patterns, whether they stem from evolutionary sampling bias or physical interaction mechanisms, remain unclear. This article provides a physics-oriented explanation for these observations, specifically concentrating on pure parallel β-sheet topologies. The insights gained from this research are crucial in enhancing our understanding of the relationship between protein sequences and structures, and are expected to contribute significantly to the de novo design of new proteins. [ABSTRACT FROM AUTHOR]

With recent methodological advances in the field of computational protein design, in particular those based on deep learning, there is an increasing need for frameworks that allow for coherent, direct integration of different models and objective functions into the generative design process. Here we demonstrate how evolutionary multiobjective optimization techniques can be adapted to provide such an approach. With the established Non-dominated Sorting Genetic Algorithm II (NSGA-II) as the optimization framework, we use AlphaFold2 and ProteinMPNN confidence metrics to define the objective space, and a mutation operator composed of ESM-1v and ProteinMPNN to rank and then redesign the least favorable positions. Using the two-state design problem of the foldswitching protein RfaH as an in-depth case study, and PapD and calmodulin as examples of higher-dimensional design problems, we show that the evolutionary multiobjective optimization approach leads to significant reduction in the bias and variance in RfaH native sequence recovery, compared to a direct application of ProteinMPNN. We suggest that this improvement is due to three factors: (i) the use of an informative mutation operator that accelerates the sequence space exploration, (ii) the parallel, iterative design process inherent to the genetic algorithm that improves upon the ProteinMPNN autoregressive sequence decoding scheme, and (iii) the explicit approximation of the Pareto front that leads to optimal design candidates representing diverse tradeoff conditions. We anticipate this approach to be readily adaptable to different models and broadly relevant for protein design tasks with complex specifications. Author summary: Proteins are the fundamental building blocks of life, and engineering them has broad applications in medicine and biotechnology. Computational methods that seek to model and predict the protein sequence-structure-function relationship have seen significant advancement from the recent development in deep learning. As more models become available, it remains an open question how to effectively combine them into a coherent computational design approach. One approach is to perform computational design with one model, and filter the design candidates with the others. In this work, we demonstrate how an optimization algorithm inspired by evolution can be adapted to provide an alternative framework that outperforms the post hoc filtering approach, especially for problems with multiple competing design specifications. Such a framework may enable researchers to more effectively integrate the strengths of different modeling approaches to tackle more complex design problems. [ABSTRACT FROM AUTHOR]

The practical application of new single molecule protein sequencing (SMPS) technologies requires accurate estimates of their associated sequencing error rates. Here, we describe the development and application of two distinct parameter estimation methods for analyzing SMPS reads produced by fluorosequencing. A Hidden Markov Model (HMM) based approach, extends whatprot, where we previously used HMMs for SMPS peptide-read matching. This extension offers a principled approach for estimating key parameters for fluorosequencing experiments, including missed amino acid cleavages, dye loss, and peptide detachment. Specifically, we adapted the Baum-Welch algorithm, a standard technique to estimate transition probabilities for an HMM using expectation maximization, but modified here to estimate a small number of parameter values directly rather than estimating every transition probability independently. We demonstrate a high degree of accuracy on simulated data, but on experimental datasets, we observed that the model needed to be augmented with an additional error type, N-terminal blocking. This, in combination with data pre-processing, results in reasonable parameterizations of experimental datasets that agree with controlled experimental perturbations. A second independent implementation using a hybrid of DIRECT and Powell's method to reduce the root mean squared error (RMSE) between simulations and the real dataset was also developed. We compare these methods on both simulated and real data, finding that our Baum-Welch based approach outperforms DIRECT and Powell's method by most, but not all, criteria. Although some discrepancies between the results exist, we also find that both approaches provide similar error rate estimates from experimental single molecule fluorosequencing datasets. Author summary: Diverse new technologies are being developed for single-molecule protein sequencing, capable of identifying and quantifying mixtures of proteins at the level of individual molecules. There are many biochemical challenges intrinsic to high-throughput studies of proteins at such high sensitivity arising from their heterogeneous chemistries, sizes, and abundances. Beyond these challenges, the technologies themselves involve complex multi-step analytical processes. Thus, in developing and optimizing these technologies, it is important to consider the accuracy of each step and to have reliable approaches for estimating these accuracies. We focus on one particular single-molecule sequencing technology known as flourosequencing. We report and validate two methods for simultaneously determining the error-rates of each of the various steps of the fluorosequencing process. These new error estimation techniques will help researchers to better interpret the effects of changes to the chemistry and sample preparation used in fluorosequencing so that these steps can be improved. Further, more accurate determination of error rates will aid in the creation of better tools for the interpretation of this data. [ABSTRACT FROM AUTHOR]

AlphaFold2 is an Artificial Intelligence-based program developed to predict the 3D structure of proteins given only their amino acid sequence at atomic resolution. Due to the accuracy and efficiency at which AlphaFold2 can generate 3D structure predictions and its widespread adoption into various aspects of biochemical research, the technique of protein structure prediction should be considered for incorporation into the undergraduate biochemistry curriculum. A module for introducing AlphaFold2 into a senior-level biochemistry laboratory classroom was developed. The module's focus was to have students predict the structures of proteins from the MPOX 22 global outbreak virus isolate genome, which had no structures elucidated at that time. The goal of this study was to both determine the impact the module had on students and to develop a framework for introducing AlphaFold2 into the undergraduate curriculum so that instructors for biochemistry courses, regardless of their background in bioinformatics, could adapt the module into their classrooms. Author summary: AlphaFold2 is software that combines sequence similarity and structure templating with the power of Artificial Intelligence (AI) to bridge the connection between the primary protein structure (amino acid sequence) and higher-level 3D structure (secondary, tertiary, and quaternary). AlphaFold2's impressive and easily accessible nature makes it a bioinformatics tool that has been seeing a wide range of applications in biochemical research. Given this large-scale application, we examined whether Alphafold2 could be integrated into an undergraduate curriculum. We developed a novel module for a senior-level undergraduate biochemistry laboratory class. Our goal was to lay a solid foundation for other undergraduate instructors to be able to adapt this module to fit their classroom needs. While we implemented and ran all predictions on an internal university computing cluster, we recommend ColabFold for those instructors who do not have access to large-scale computational clusters or whose internal clusters cannot scale to their classroom sizes. We have outlined 3 metrics to be quantitatively investigated in the module to give both instructors and students metrics to evaluate model confidence. We have also included a template worksheet, lecture slides, and example scripts to enable instructors to rapidly develop a similar module. We hope that more departments and programs will integrate AlphaFold2 into their undergraduate curriculums, giving students a highly in-demand skill to prepare them for their transition into a career or graduate school. [ABSTRACT FROM AUTHOR]

The activities of most enzymes and drugs depend on interactions between proteins and small molecules. Accurate prediction of these interactions could greatly accelerate pharmaceutical and biotechnological research. Current machine learning models designed for this task have a limited ability to generalize beyond the proteins used for training. This limitation is likely due to a lack of information exchange between the protein and the small molecule during the generation of the required numerical representations. Here, we introduce ProSmith, a machine learning framework that employs a multimodal Transformer Network to simultaneously process protein amino acid sequences and small molecule strings in the same input. This approach facilitates the exchange of all relevant information between the two molecule types during the computation of their numerical representations, allowing the model to account for their structural and functional interactions. Our final model combines gradient boosting predictions based on the resulting multimodal Transformer Network with independent predictions based on separate deep learning representations of the proteins and small molecules. The resulting predictions outperform recently published state-of-the-art models for predicting protein-small molecule interactions across three diverse tasks: predicting kinase inhibitions; inferring potential substrates for enzymes; and predicting Michaelis constants KM. The Python code provided can be used to easily implement and improve machine learning predictions involving arbitrary protein-small molecule interactions. Author summary: Understanding how proteins interact with small molecules, such as drugs, is critical to advancing medical, biological, and biotechnological research. Our work introduces ProSmith, a machine learning framework that improves the prediction of protein-small molecule interactions. Protein-small molecule interactions can be predicted by using numerical representations of proteins and small molecules as input to machine learning prediction models. Previous methods typically generated separate numerical representations for the proteins and small molecules without considering their interactions. ProSmith, however, combines both protein sequence and small molecule structural information in the input of a single multimodal Transformer Network to generate a joint numerical representation. Unlike previous methods, this allows for a comprehensive exchange of information between protein and small molecule, capturing the complex relationships and interactions between these two types of molecules. ProSmith successfully predicts several biological interactions, including kinase inhibitions, potential enzyme-substrate pairs, and enzyme kinetic parameters KM. We provide Python code that can be easily adapted to improve predictions for any protein-small molecule interaction. [ABSTRACT FROM AUTHOR]

The design of proteins with specific tasks is a major challenge in molecular biology with important diagnostic and therapeutic applications. High-throughput screening methods have been developed to systematically evaluate protein activity, but only a small fraction of possible protein variants can be tested using these techniques. Computational models that explore the sequence space in-silico to identify the fittest molecules for a given function are needed to overcome this limitation. In this article, we propose AnnealDCA, a machine-learning framework to learn the protein fitness landscape from sequencing data derived from a broad range of experiments that use selection and sequencing to quantify protein activity. We demonstrate the effectiveness of our method by applying it to antibody Rep-Seq data of immunized mice and screening experiments, assessing the quality of the fitness landscape reconstructions. Our method can be applied to several experimental cases where a population of protein variants undergoes various rounds of selection and sequencing, without relying on the computation of variants enrichment ratios, and thus can be used even in cases of disjoint sequence samples. Author summary: Advances in sequencing techniques have recently generated an explosion of protein sequence data. This represents an opportunity for scientists to develop theoretical and computational methods that can extract relevant biological information from these data samples. In this perspective, machine learning methods are proving to be particularly effective in the biological context. Since the majority of the accessible protein sequences are not-annotated, i.e. no information about the functional properties is known, unsupervised machine learning methods are particularly suited to tackle such raw sequence data. Here, we propose an unsupervised inference method which is meant to be applied to protein sequence data generated by an evolutionary process, whether it takes place in a controlled experimental framework or in-vivo. The method is devised to be simple enough to be applied to a plethora of different experimental setups, at the same time modeling the fundamental features of the dynamical processes underlying data generation. The ultimate goal of the method is to provide a sequence-fitness mapping that goes beyond the experimentally assessed sequence space, so to assign a quantitative functional score to each possible protein variant. The accurate knowledge of this mapping is key for several biological applications, such as biomolecule design and engineering, diagnostic and therapeutic treatments, and vaccine development. [ABSTRACT FROM AUTHOR]

One of the challenges in a viral pandemic is the emergence of novel variants with different phenotypical characteristics. An ability to forecast future viral individuals at the sequence level enables advance preparation by characterizing the sequences and closing vulnerabilities in current preventative and therapeutic methods. In this article, we explore, in the context of a viral pandemic, the problem of generating complete instances of undiscovered viral protein sequences, which have a high likelihood of being discovered in the future using protein language models. Current approaches to training these models fit model parameters to a known sequence set, which does not suit pandemic forecasting as future sequences differ from known sequences in some respects. To address this, we develop a novel method, called PandoGen, to train protein language models towards the pandemic protein forecasting task. PandoGen combines techniques such as synthetic data generation, conditional sequence generation, and reward-based learning, enabling the model to forecast future sequences, with a high propensity to spread. Applying our method to modeling the SARS-CoV-2 Spike protein sequence, we find empirically that our model forecasts twice as many novel sequences with five times the case counts compared to a model that is 30× larger. Our method forecasts unseen lineages months in advance, whereas models 4× and 30× larger forecast almost no new lineages. When trained on data available up to a month before the onset of important Variants of Concern, our method consistently forecasts sequences belonging to those variants within tight sequence budgets. Author summary: Viral protein sequences play a pivotal role in the spread of a pandemic. As the virus evolves, so do the viral proteins, increasing the potency of the virus. Knowledge of future viral protein sequences can be invaluable because it allows us to test the efficacy of preventative and treatment methods against future changes to the virus, and tailor them to such changes early. We attempt to forecast viral proteins ahead of time. Making such predictions is very challenging and complex because the prediction target is a sequence with thousands of positions, and a single mis-predicted sequence position may invalidate the entire prediction. Also, as the virus continues to evolve, the data available to train models becomes obsolete. Addressing these challenges, we create a novel approach to train models of the SARS-CoV-2 Spike protein, that are especially tailored to forecasting future sequences. Models trained using this approach outperform existing approaches in their effectiveness. In addition, our method can train models to forecast important pandemic variants ahead of time. [ABSTRACT FROM AUTHOR]

Recent advances in deep learning have significantly improved the ability to infer protein sequences directly from protein structures for the fix-backbone design. The methods have evolved from the early use of multi-layer perceptrons to convolutional neural networks, transformers, and graph neural networks (GNN). However, the conventional approach of constructing K-nearest-neighbors (KNN) graph for GNN has limited the utilization of edge information, which plays a critical role in network performance. Here we introduced SPIN-CGNN based on protein contact maps for nearest neighbors. Together with auxiliary edge updates and selective kernels, we found that SPIN-CGNN provided a comparable performance in refolding ability by AlphaFold2 to the current state-of-the-art techniques but a significant improvement over them in term of sequence recovery, perplexity, deviation from amino-acid compositions of native sequences, conservation of hydrophobic positions, and low complexity regions, according to the test by unseen structures, "hallucinated" structures and diffusion models. Results suggest that low complexity regions in the sequences designed by deep learning, for generated structures in particular, remain to be improved, when compared to the native sequences. Author summary: We proposed SPIN-CGNN, a deep learning-based method for fixed backbone protein design. Our studies showed that introducing contact map-based graph construction, second-order edge updates, and selective kernels cumulatively improved the performance over existing methods in native and generated structures according to multiple measures, including amino-acid compositions, amino-acid substitutions, low complexity regions, and conservations of hydrophobic/hydrophilic positions for specific evaluation of fix backbone protein design methods. [ABSTRACT FROM AUTHOR]

T cells rely on their T cell receptors (TCRs) to discern foreign antigens presented by human leukocyte antigen (HLA) proteins. The TCRs of an individual contain a record of this individual's past immune activities, such as immune response to infections or vaccines. Mining the TCR data may recover useful information or biomarkers for immune related diseases or conditions. Some TCRs are observed only in the individuals with certain HLA alleles, and thus characterizing TCRs requires a thorough understanding of TCR-HLA associations. The extensive diversity of HLA alleles and the rareness of some HLA alleles present a formidable challenge for this task. Existing methods either treat HLA as a categorical variable or represent an HLA by its alphanumeric name, and have limited ability to generalize to the HLAs that are not seen in the training process. To address this challenge, we propose a neural network-based method named Deep learning Prediction of TCR-HLA association (DePTH) to predict TCR-HLA associations based on their amino acid sequences. We demonstrate that DePTH is capable of making reasonable predictions for TCR-HLA associations, even when neither the HLA nor the TCR have been included in the training dataset. Furthermore, we establish that DePTH can be used to quantify the functional similarities among HLA alleles, and that these HLA similarities are associated with the survival outcomes of cancer patients who received immune checkpoint blockade treatments. Author summary: T cells are critical components of the human immune system. A T cell receptor (TCR) is a protein complex found on the surface of T cells, and it determines the antigens that a T cell recognizes. The TCR repertoire, which is the collection of all the TCRs within an individual, provides rich information about the history and current activities of the immune system. A TCR-antigen interaction is mediated by a protein called human leukocyte antigen (HLA). HLA genes are highly polymorphic in the human population, contributing to the variation of human immune response to different types of antigens. Studying the associations between TCRs and HLAs can help us identify functional TCRs given the HLAs of an individual. To this end, we develop a deep learning method to predict TCR-HLA associations based on their amino acid sequences. This method allows us to borrow information across different HLA genes. We demonstrate that the predictions of our model can be used to quantify the functional similarities of HLA alleles and such similarities are associated with cancer patients' survival outcome. [ABSTRACT FROM AUTHOR]

We present here an approach to protein design that combines (i) scarce functional information such as experimental data (ii) evolutionary information learned from a natural sequence variants and (iii) physics-grounded modeling. Using a Restricted Boltzmann Machine (RBM), we learn a sequence model of a protein family. We use semi-supervision to leverage available functional information during the RBM training. We then propose a strategy to explore the protein representation space that can be informed by external models such as an empirical force-field method (FoldX). Our approach is applied to a domain of the Cas9 protein responsible for recognition of a short DNA motif. We experimentally assess the functionality of 71 variants generated to explore a range of RBM and FoldX energies. Sequences with as many as 50 differences (20% of the protein domain) to the wild-type retained functionality. Overall, 21/71 sequences designed with our method were functional. Interestingly, 6/71 sequences showed an improved activity in comparison with the original wild-type protein sequence. These results demonstrate the interest in further exploring the synergies between machine-learning of protein sequence representations and physics grounded modeling strategies informed by structural information. Author summary: Proteins are essential molecules in all living organisms, with their function largely determined by their sequence. Modifying a protein's sequence to achieve a desired function remains a challenging endeavor, requiring careful consideration of factors such as the stability of the structure and interactions with molecular partners. In this study, we devised a protein design method that combines insights from experimental data, natural variation in protein sequences, and physics-based predictions. This approach provides a reliable and interpretable means of altering a protein's sequence while maintaining its functionality. We applied our technique to a domain of the Cas9 protein, a key component in the CRISPR gene editing system. Our results demonstrate the possibility of generating functional protein domains with over 20% of their sequence modified. These findings underscore how the integration of diverse sources of information in a unified design process enhances the quality of engineered proteins. This advancement holds promise for creating valuable protein variants for applications in drug development and various industries. [ABSTRACT FROM AUTHOR]

The adaptive immune system employs an array of receptors designed to respond with high specificity to pathogens or molecular aberrations faced by the host organism. Binding of these receptors to molecular fragments—collectively referred to as antigens—initiates immune responses. These antigenic targets are recognized in their native state on the surfaces of pathogens by antibodies, whereas T cell receptors (TCR) recognize processed antigens as short peptides, presented on major histocompatibility complex (MHC) molecules. Recent research has led to a wealth of immune repertoire data that are key to interrogating the nature of these molecular interactions. However, existing tools for the analysis of these large datasets typically focus on molecular sets of a single type, forcing researchers to separately analyze strongly coupled sequences of interacting molecules. Here, we introduce a software package for the integrated analysis of immune repertoire data, capable of identifying distinct biophysical differences in isolated TCR, MHC, peptide, antibody, and antigen sequence data. This integrated analytical approach allows for direct comparisons across immune repertoire subsets and provides a starting point for the identification of key interaction hotspots in complementary receptor-antigen pairs. The software (AIMS—Automated Immune Molecule Separator) is freely available as an open access package in GUI or command-line form. Author summary: Over the past decade, the success of immunotherapeutics coupled with the declining costs of sequencing have stimulated a near exponential growth in the identification of novel T cell receptor, peptide, and antibody sequences for use in combating disease and dysregulation. With these new datasets freely available to researchers, a wealth of analytical tools have been created for standardized data analysis. However, these tools are largely fragmented, capable of processing only singular molecular species, likewise generating fragmented interpretations of complex adaptive immune environments. In this manuscript, we outline the capabilities of a new analytical tool, the AIMS: Automated Immune Molecule Separator software, designed for the uniform analysis of all adaptive immune molecules. AIMS accomplishes this cross-receptor compatibility using an amino acid sequence encoding approach that captures key biophysical properties without requiring explicit experimental structural data. The software can be extended to non-immune molecules, making AIMS a widely applicable platform for the broader analysis of protein-protein interactions. [ABSTRACT FROM AUTHOR]

Gene expression is the synthesis of proteins from the information encoded on DNA. One of the two main steps of gene expression is the translation of messenger RNA (mRNA) into polypeptide sequences of amino acids. Here, by taking into account mRNA degradation, we model the motion of ribosomes along mRNA with a ballistic model where particles advance along a filament without excluded volume interactions. Unidirectional models of transport have previously been used to fit the average density of ribosomes obtained by the experimental ribo-sequencing (Ribo-seq) technique in order to obtain the kinetic rates. The degradation rate is not, however, accounted for and experimental data from different experiments are needed to have enough parameters for the fit. Here, we propose an entirely novel experimental setup and theoretical framework consisting in splitting the mRNAs into categories depending on the number of ribosomes from one to four. We solve analytically the ballistic model for a fixed number of ribosomes per mRNA, study the different regimes of degradation, and propose a criterion for the quality of the inverse fit. The proposed method provides a high sensitivity to the mRNA degradation rate. The additional equations coming from using the monosome (single ribosome) and polysome (arbitrary number) ribo-seq profiles enable us to determine all the kinetic rates in terms of the experimentally accessible mRNA degradation rate. Author summary: Genetic translation, involving the transport of ribosomes, is one of the main steps of genetic expression. The molecular mechanisms that regulate ribosome motion along messenger RNA (mRNA) remain to be elucidated. Ribo-sequencing experiments are now capable of providing access to the gene-dependent density of ribosomes along mRNA. The physical modeling of ribosome motion is carried out using 1D transport models, such as TASEP-like ones, which enables one to obtain the kinetic (initiation and elongation) rates from the Ribo-seq densities. Currently used methods, however, do not allow access to mRNA degradation rates and previous experiments lack input to perform a self-consistent quantitative analysis. Here, we propose a new experimental and modeling protocol to obtain the kinetic rates (including mRNA degradation) with better accuracy. Ribo-seq is performed after splitting the mRNA population into groups that depend on the number of translating ribosomes (from 1 to 4). We thus obtain four Ribo-seq profiles per gene properly normalized without requiring independent experiments. We show that these low ribosome number profiles display a higher sensitivity to mRNA degradation than the full polysome profile and discuss the biological translation regimes of several organisms. [ABSTRACT FROM AUTHOR]

We present a machine learning-based interpretive framework (whatprot) for analyzing single molecule protein sequencing data produced by fluorosequencing, a recently developed proteomics technology that determines sparse amino acid sequences for many individual peptide molecules in a highly parallelized fashion. Whatprot uses Hidden Markov Models (HMMs) to represent the states of each peptide undergoing the various chemical processes during fluorosequencing, and applies these in a Bayesian classifier, in combination with pre-filtering by a k-Nearest Neighbors (kNN) classifier trained on large volumes of simulated fluorosequencing data. We have found that by combining the HMM based Bayesian classifier with the kNN pre-filter, we are able to retain the benefits of both, achieving both tractable runtimes and acceptable precision and recall for identifying peptides and their parent proteins from complex mixtures, outperforming the capabilities of either classifier on its own. Whatprot's hybrid kNN-HMM approach enables the efficient interpretation of fluorosequencing data using a full proteome reference database and should now also enable improved sequencing error rate estimates. Author summary: Scientists often wish to know which proteins, and at what quantities, are present in a sample. The field of proteomics offers a number of technologies that aid in this, such as tandem mass spectrometry and immunoassays, that provide different tradeoffs between sensitivity, throughput, and generality. One new technology, known as fluorosequencing, detects and provides partial sequences for individual peptide or protein molecules from a sample in a highly parallelized fashion. However, as only partial sequences are measured, the resulting sequencing reads must be matched to a reference database of possible proteins, such as might be obtained from the human genome. We describe a suitable computer algorithm for performing this matching of fluorosequencing reads to a reference database while accounting for the most prevalent types of sequencing errors. We detail its performance and implementation, and describe a number of uncommon algorithmic improvements and approximations which allow this approach to scale to classification against the whole human proteome. The resulting software, known as whatprot, allows researchers to interpret fluorosequencing reads and better apply this emergent single molecule protein sequencing technology. [ABSTRACT FROM AUTHOR]

Protein design is a technique to engineer proteins by permuting amino acids in the sequence to obtain novel functionalities. However, exploring all possible combinations of amino acids is generally impossible due to the exponential growth of possibilities with the number of designable sites. The present work introduces circuits implementing a pure quantum approach, Grover's algorithm, to solve protein design problems. Our algorithms can adjust to implement any custom pair-wise energy tables and protein structure models. Moreover, the algorithm's oracle is designed to consist of only adder functions. Quantum computer simulators validate the practicality of our circuits, containing up to 234 qubits. However, a smaller circuit is implemented on real quantum devices. Our results show that using O(N) iterations, the circuits find the correct results among all N possibilities, providing the expected quadratic speed up of Grover's algorithm over classical methods (i.e., O(N)). Author summary: Protein design aims to create novel proteins or enhance the functionality of existing proteins by tweaking their sequences through permuting amino acids. The number of possible configurations, N, grows exponentially as a function of the number of designable sites (s), i.e., N = As, where A is the number of different amino acids (A = 20 for canonical amino acids). The classical computation methods require O(N)) queries to search and find the low-energy configurations among N possible sequences. Searching among these possibilities becomes unattainable for large proteins, forcing the classical approaches to use sampling methods. Alternatively, quantum computing can promise quadratic speed-up in searching for answers in an unorganized list by employing Grover's algorithm. Our work shows the implementation of this algorithm at the circuit level to solve protein design problems. We first focus on lattice model-like systems and then improve them to more realistic models (change in the energy as a function of distances). Our algorithms can implement various custom pair-wise energy tables and any protein structure models. We have used quantum computer simulators to validate the practicality of our circuits which require up to 234 qubits. We have also implemented a simple version of our circuits on real quantum devices. Our results show that our circuits provide the expected quadratic speed-up of Grover's algorithm. [ABSTRACT FROM AUTHOR]

Generating and analyzing overlapping peptides through multienzymatic digestion is an efficient procedure for de novo protein using from bottom-up mass spectrometry (MS). Despite improved instrumentation and software, de novo MS data analysis remains challenging. In recent years, deep learning models have represented a performance breakthrough. Incorporating that technology into de novo protein sequencing workflows require machine-learning models capable of handling highly diverse MS data. In this study, we analyzed the requirements for assembling such generalizable deep learning models by systemcally varying the composition and size of the training set. We assessed the generated models' performances using two test sets composed of peptides originating from the multienzyme digestion of samples from various species. The peptide recall values on the test sets showed that the deep learning models generated from a collection of highly N- and C-termini diverse peptides generalized 76% more over the termini-restricted ones. Moreover, expanding the training set's size by adding peptides from the multienzymatic digestion with five proteases of several species samples led to a 2–3 fold generalizability gain. Furthermore, we tested the applicability of these multienzyme deep learning (MEM) models by fully de novo sequencing the heavy and light monomeric chains of five commercial antibodies (mAbs). MEMs extracted over 10000 matching and overlapped peptides across six different proteases mAb samples, achieving a 100% sequence coverage for 8 of the ten polypeptide chains. We foretell that the MEMs' proven improvements to de novo analysis will positively impact several applications, such as analyzing samples of high complexity, unknown nature, or the peptidomics field. Author summary: In recent years, the application of deep learning represented a breakthrough in the mass spectrometry (MS) field by improving the assignment of the correct sequence of amino acids from observable MS spectra without prior knowledge, also known as de novo MS-based peptide sequencing. However, like other modern neural networks, models do not generalize well enough as they perform poorly on highly varied N- and C-termini peptide test sets. To mitigate this generalizability problem, we conducted a systematic investigation to uncover the requirements for building generalized models and boosting the performance on the MS-based de novo peptide sequencing task. Several experiments confirmed that the training set's peptide diversity directly impacts the resulting model's generalizability. Data showed that the best models were the multienzyme models (MEMs), i.e., models trained from a compendium of highly diverse peptides, such as the one generated from digesting a broad of species samples with a group of proteases. The applicability of these MEMs was later established by fully de novo sequencing 8 of the ten polypeptide chains of five commercial antibodies and extracting over 10000 proving peptides. [ABSTRACT FROM AUTHOR]

Constraint-based modeling is a powerful framework for studying cellular metabolism, with applications ranging from predicting growth rates and optimizing production of high value metabolites to identifying enzymes in pathogens that may be targeted for therapeutic interventions. Results from modeling experiments can be affected at least in part by the quality of the metabolic models used. Reconstructing a metabolic network manually can produce a high-quality metabolic model but is a time-consuming task. At the same time, current methods for automating the process typically transfer metabolic function based on sequence similarity, a process known to produce many false positives. We created Architect, a pipeline for automatic metabolic model reconstruction from protein sequences. First, it performs enzyme annotation through an ensemble approach, whereby a likelihood score is computed for an EC prediction based on predictions from existing tools; for this step, our method shows both increased precision and recall compared to individual tools. Next, Architect uses these annotations to construct a high-quality metabolic network which is then gap-filled based on likelihood scores from the ensemble approach. The resulting metabolic model is output in SBML format, suitable for constraints-based analyses. Through comparisons of enzyme annotations and curated metabolic models, we demonstrate improved performance of Architect over other state-of-the-art tools, notably with higher precision and recall on the eukaryote C. elegans and when compared to UniProt annotations in two bacterial species. Code for Architect is available at https://github.com/ParkinsonLab/Architect. For ease-of-use, Architect can be readily set up and utilized using its Docker image, maintained on Docker Hub. Author summary: An organism's growth and survival are largely guided by its ability to synthesize crucial metabolites like amino acids and ribonucleotides from such compounds as water, glucose and nitrogenous molecules like ammonia. Accurate knowledge of such biochemical reactions—catalyzed by enzymes encoded within the genome—can advance our understanding of disease drivers as well as guide attempts at engineering strains of bacteria with desired metabolic capacities. While biochemical experiments can accurately characterize metabolism, these are time-consuming at genome-scale. Instead, genome-scale metabolic models can be computationally built then iteratively refined through comparisons of in silico simulation results and biochemical experiments. Here, we describe Architect, a method for automatic enzyme annotation and metabolic model reconstruction. Our tool leverages the strengths of existing enzyme annotation tools to predict the biochemical capacities of an organism and then uses these predictions to build a simulation-ready metabolic model. We find that Architect produces more accurate enzyme annotations than the individual tools, as well as higher-quality metabolic models compared to other automatic metabolic model reconstruction tools. We provide Architect to the metabolic modelling community in the hope that it may facilitate the transition from knowing an organism's encoded sequences to an understanding of its metabolic capacities. [ABSTRACT FROM AUTHOR]

A major challenge to the characterization of intrinsically disordered regions (IDRs), which are widespread in the proteome, but relatively poorly understood, is the identification of molecular features that mediate functions of these regions, such as short motifs, amino acid repeats and physicochemical properties. Here, we introduce a proteome-scale feature discovery approach for IDRs. Our approach, which we call "reverse homology", exploits the principle that important functional features are conserved over evolution. We use this as a contrastive learning signal for deep learning: given a set of homologous IDRs, the neural network has to correctly choose a held-out homolog from another set of IDRs sampled randomly from the proteome. We pair reverse homology with a simple architecture and standard interpretation techniques, and show that the network learns conserved features of IDRs that can be interpreted as motifs, repeats, or bulk features like charge or amino acid propensities. We also show that our model can be used to produce visualizations of what residues and regions are most important to IDR function, generating hypotheses for uncharacterized IDRs. Our results suggest that feature discovery using unsupervised neural networks is a promising avenue to gain systematic insight into poorly understood protein sequences. Author summary: Intrinsically disordered regions (IDRs) are widespread in proteins but are poorly understood on a systematic level because they evolve too rapidly for classic bioinformatics methods to be effective. We designed a neural network that learns what features (for example, electrostatic charge, or the presence of certain motifs) might be important to the function of IDRs, even when we don't have prior knowledge of function. Our neural network learns by exploiting principles of evolution. Important features tend to be conserved over species, so guessing what sequences evolved from the same common ancestor helps the neural network identify these features. Importantly, training a neural network this way can be defined as a fully automatic operation, so no manual effort is required. After our neural network is trained, we can apply interpretation techniques to understand what kinds of features are important to IDRs globally in the proteome, and to form hypotheses about specific IDRs. We show that many of the features our neural network learns are consistent with features we already know to be important to IDRs. We hope that our neural network can be applied to help biologists form hypotheses about poorly characterized IDRs. [ABSTRACT FROM AUTHOR]

Many different types of generative models for protein sequences have been proposed in literature. Their uses include the prediction of mutational effects, protein design and the prediction of structural properties. Neural network (NN) architectures have shown great performances, commonly attributed to the capacity to extract non-trivial higher-order interactions from the data. In this work, we analyze two different NN models and assess how close they are to simple pairwise distributions, which have been used in the past for similar problems. We present an approach for extracting pairwise models from more complex ones using an energy-based modeling framework. We show that for the tested models the extracted pairwise models can replicate the energies of the original models and are also close in performance in tasks like mutational effect prediction. In addition, we show that even simpler, factorized models often come close in performance to the original models. Author summary: Complex neural networks trained on large biological datasets have recently shown powerful capabilites in tasks like the prediction of protein structure, assessing the effect of mutations on the fitness of proteins and even designing completely novel proteins with desired characteristics. The enthralling prospect of leveraging these advances in fields like medicine and synthetic biology has created a large amount of interest in academic research and industry. The connected question of what biological insights these methods actually gain during training has, however, received less attention. In this work, we systematically investigate in how far neural networks capture information that could not be captured by simpler models. To this end, we develop a method to train simpler models to imitate more complex models, and compare their performance to the original neural network models. Surprisingly, we find that the simpler models thus trained often perform on par with the neural networks, while having a considerably easier structure. This highlights the importance of finding ways to interpret the predictions of neural networks in these fields, which could inform the creation of better models, improve methods for their assessment and ultimately also increase our understanding of the underlying biology. [ABSTRACT FROM AUTHOR]

Recent advances in experimental and computational protein structure determination have provided access to high-quality structures for most human proteins and mutants thereof. However, linking changes in structure in protein mutants to functional impact remains an active area of method development. If successful, such methods can ultimately assist physicians in taking appropriate treatment decisions. This work presents three artificial neural network (ANN)-based predictive models that classify four key functional parameters of KCNQ1 variants as normal or dysfunctional using PSSM-based evolutionary and/or biophysical descriptors. Recent advances in predicting protein structure and variant properties with artificial intelligence (AI) rely heavily on the availability of evolutionary features and thus fail to directly assess the biophysical underpinnings of a change in structure and/or function. The central goal of this work was to develop an ANN model based on structure and physiochemical properties of KCNQ1 potassium channels that performs comparably or better than algorithms using only on PSSM-based evolutionary features. These biophysical features highlight the structure-function relationships that govern protein stability, function, and regulation. The input sensitivity algorithm incorporates the roles of hydrophobicity, polarizability, and functional densities on key functional parameters of the KCNQ1 channel. Inclusion of the biophysical features outperforms exclusive use of PSSM-based evolutionary features in predicting activation voltage dependence and deactivation time. As AI is increasing applied to problems in biology, biophysical understanding will be critical with respect to 'explainable AI', i.e., understanding the relation of sequence, structure, and function of proteins. Our model is available at www.kcnq1predict.org. Author summary: Heartbeat is maintained by electrical impulses generated by ion-conducting channel proteins in the heart such as the KCNQ1 potassium channel. Pathogenic variants in KCNQ1 can lead to channel loss-of-function and predisposition to fatal life-threatening irregularities of heart rhythm (arrhythmia). Machine learning methods that can predict the outcome of a mutation on KCNQ1 structure and function would be of great value in helping to assess the risk of a heart rhythm disorder. Recently, machine learning has made great progress in predicting the structures of proteins from their sequences. However, there are limited studies that link the effect of a mutation and change in protein structure with its function. This work presents the development of neural network models designed to predict mutation-induced changes in KCNQ1 functional parameters such as peak current density and voltage dependence of activation. We compare the predictive ability of features extracted from sequence, structure, and physicochemical properties of KCNQ1. Moreover, input sensitivity analysis connects biophysical features with specific functional parameters that provides insight into underlying molecular mechanisms for KCNQ1 channels. The best performing neural network model is publicly available as a webserver, called Q1VarPredBio, that delivers predictions about the functional phenotype of KCNQ1 variants. [ABSTRACT FROM AUTHOR]

One of the unique traits of membrane proteins is that a significant fraction of their hydrophobic amino acids is exposed to the hydrophobic core of lipid bilayers rather than being embedded in the protein interior, which is often not explicitly considered in the protein structure and function predictions. Here, we propose a characteristic and predictive quantity, the membrane contact probability (MCP), to describe the likelihood of the amino acids of a given sequence being in direct contact with the acyl chains of lipid molecules. We show that MCP is complementary to solvent accessibility in characterizing the outer surface of membrane proteins, and it can be predicted for any given sequence with a machine learning-based method by utilizing a training dataset extracted from MemProtMD, a database generated from molecular dynamics simulations for the membrane proteins with a known structure. As the first of many potential applications, we demonstrate that MCP can be used to systematically improve the prediction precision of the protein contact maps and structures. Author summary: The distribution of residues on protein surfaces is largely determined by the surrounding environment. For soluble proteins, most of the residues on the outer surface are hydrophilic, and people use the quantity "solvent accessibility" to describe and predict these surface residues. In contrast, for membrane proteins that are embedded in a lipid bilayer, many of their surface residues are hydrophobic and membrane-contacting, but there is yet a widely-accepted quantity for the description or prediction of this characteristic property. Here, we propose a new quantity termed "membrane contact probability (MCP)", which can be used to describe and predict the membrane-contacting surface residues of proteins. We also propose a machine learning-based method to predict MCP from protein sequences, utilizing the dataset generated by physics-based computer simulations. We demonstrate that a quantity such as MCP is helpful for protein structure prediction, and we believe that it will find broad applications in the structure and function studies of membrane proteins. [ABSTRACT FROM AUTHOR]

All proteomes contain both proteins and polypeptide segments that don't form a defined three-dimensional structure yet are biologically active—called intrinsically disordered proteins and regions (IDPs and IDRs). Most of these IDPs/IDRs lack useful functional annotation limiting our understanding of their importance for organism fitness. Here we characterized IDRs using protein sequence annotations of functional sites and regions available in the UniProt knowledgebase ("UniProt features": active site, ligand-binding pocket, regions mediating protein-protein interactions, etc.). By measuring the statistical enrichment of twenty-five UniProt features in 981 IDRs of 561 human proteins, we identified eight features that are commonly located in IDRs. We then collected the genetic variant data from the general population and patient-based databases and evaluated the prevalence of population and pathogenic variations in IDPs/IDRs. We observed that some IDRs tolerate 2 to 12-times more single amino acid-substituting missense mutations than synonymous changes in the general population. However, we also found that 37% of all germline pathogenic mutations are located in disordered regions of 96 proteins. Based on the observed-to-expected frequency of mutations, we categorized 34 IDRs in 20 proteins (DDX3X, KIT, RB1, etc.) as intolerant to mutation. Finally, using statistical analysis and a machine learning approach, we demonstrate that mutation-intolerant IDRs carry a distinct signature of functional features. Our study presents a novel approach to assign functional importance to IDRs by leveraging the wealth of available genetic data, which will aid in a deeper understating of the role of IDRs in biological processes and disease mechanisms. Author summary: Intrinsically disordered regions (IDRs) in proteins are typically not considered to be functionally as important as the structured parts. However, it is becoming evident that both structured and disordered regions are essential for the repertoire of protein functions. Nevertheless, most of these largely flexible and functionally dynamic protein regions remain uncharacterized. Nevertheless, most of these largely flexible and functionally dynamic protein regions remain uncharacterized. Here, informed by human genetic diversity (i.e., genetic variations from the general population and patients), we identified the IDRs that are more frequently mutated in patients than in relatively healthy individuals, and further show that they carry a set of characteristic functional features. This approach provides a different and effective means to identify unannotated disordered protein segments that are biologically important and lead to pathogenesis upon mutation. [ABSTRACT FROM AUTHOR]

Biocatalysis is a promising approach to sustainably synthesize pharmaceuticals, complex natural products, and commodity chemicals at scale. However, the adoption of biocatalysis is limited by our ability to select enzymes that will catalyze their natural chemical transformation on non-natural substrates. While machine learning and in silico directed evolution are well-posed for this predictive modeling challenge, efforts to date have primarily aimed to increase activity against a single known substrate, rather than to identify enzymes capable of acting on new substrates of interest. To address this need, we curate 6 different high-quality enzyme family screens from the literature that each measure multiple enzymes against multiple substrates. We compare machine learning-based compound-protein interaction (CPI) modeling approaches from the literature used for predicting drug-target interactions. Surprisingly, comparing these interaction-based models against collections of independent (single task) enzyme-only or substrate-only models reveals that current CPI approaches are incapable of learning interactions between compounds and proteins in the current family level data regime. We further validate this observation by demonstrating that our no-interaction baseline can outperform CPI-based models from the literature used to guide the discovery of kinase inhibitors. Given the high performance of non-interaction based models, we introduce a new structure-based strategy for pooling residue representations across a protein sequence. Altogether, this work motivates a principled path forward in order to build and evaluate meaningful predictive models for biocatalysis and other drug discovery applications. Author summary: Predicting interactions between compounds and proteins represents a long-standing dream of drug discovery and protein engineering. Robust models of enzyme-substrate scope would dramatically advance our ability to design synthetic routes involving enzymatic catalysis. However, the lack of standardization between compound-protein interaction studies makes it difficult to evaluate the generalizability of such models. In this work we take a critical step forward by standardizing high-quality datasets measuring enzyme-substrate interactions, outlining rigorous evaluations, and proposing a new way to integrate structural information into protein representations. In testing previous modeling approaches, we highlight a surprising inability of existing models to effectively leverage compound-protein interactions to improve generalization, which challenges a perception in the literature. This establishes future opportunities for model development and integration of enzyme-substrate scope models into computer-aided synthesis planning software. [ABSTRACT FROM AUTHOR]

MHC-I associated peptides (MAPs) play a central role in the elimination of virus-infected and neoplastic cells by CD8 T cells. However, accurately predicting the MAP repertoire remains difficult, because only a fraction of the transcriptome generates MAPs. In this study, we investigated whether codon arrangement (usage and placement) regulates MAP biogenesis. We developed an artificial neural network called Codon Arrangement MAP Predictor (CAMAP), predicting MAP presentation solely from mRNA sequences flanking the MAP-coding codons (MCCs), while excluding the MCC per se. CAMAP predictions were significantly more accurate when using original codon sequences than shuffled codon sequences which reflect amino acid usage. Furthermore, predictions were independent of mRNA expression and MAP binding affinity to MHC-I molecules and applied to several cell types and species. Combining MAP ligand scores, transcript expression level and CAMAP scores was particularly useful to increase MAP prediction accuracy. Using an in vitro assay, we showed that varying the synonymous codons in the regions flanking the MCCs (without changing the amino acid sequence) resulted in significant modulation of MAP presentation at the cell surface. Taken together, our results demonstrate the role of codon arrangement in the regulation of MAP presentation and support integration of both translational and post-translational events in predictive algorithms to ameliorate modeling of the immunopeptidome. Author summary: MHC-I associated peptides (MAPs) are small fragments of intracellular proteins presented at the surface of cells and used by the immune system to detect and eliminate cancerous or virus-infected cells. While it is theoretically possible to predict which portions of the intracellular proteins will be naturally processed by the cells to ultimately reach the surface, current methodologies have prohibitively high false discovery rates. Here we introduce an artificial neural network called Codon Arrangement MAP Predictor (CAMAP) which integrates information from mRNA-to-protein translation to other factors regulating MAP biogenesis (e.g. MAP ligand score and transcript expression levels) to improve MAP prediction accuracy. While most MAP predictive approaches focus on MAP sequences per se, CAMAP's novelty is to analyze the MAP-flanking mRNA sequences, thereby providing completely independent information for MAP prediction. We show on several datasets that the integration of CAMAP scores with other known factors involved in MAP presentation (i.e. MAP ligand score and mRNA expression) significantly improves MAP prediction accuracy, and further validate CAMAP learned features using an in-vitro assay. These findings may have major implications for the design of vaccines against cancers and viruses, and in times of pandemics could accelerate the identification of relevant MAPs of viral origins. [ABSTRACT FROM AUTHOR]

Secondary metabolites (SMs) are a vast group of compounds with different structures and properties that have been utilized as drugs, food additives, dyes, and as monomers for novel plastics. In many cases, the biosynthesis of SMs is catalysed by enzymes whose corresponding genes are co-localized in the genome in biosynthetic gene clusters (BGCs). Notably, BGCs may contain so-called gap genes, that are not involved in the biosynthesis of the SM. Current genome mining tools can identify BGCs, but they have problems with distinguishing essential genes from gap genes. This can and must be done by expensive, laborious, and time-consuming comparative genomic approaches or transcriptome analyses. In this study, we developed a method that allows semi-automated identification of essential genes in a BGC based on co-evolution analysis. To this end, the protein sequences of a BGC are blasted against a suitable proteome database. For each protein, a phylogenetic tree is created. The trees are compared by treeKO to detect co-evolution. The results of this comparison are visualized in different output formats, which are compared visually. Our results suggest that co-evolution is commonly occurring within BGCs, albeit not all, and that especially those genes that encode for enzymes of the biosynthetic pathway are co-evolutionary linked and can be identified with FunOrder. In light of the growing number of genomic data available, this will contribute to the studies of BGCs in native hosts and facilitate heterologous expression in other organisms with the aim of the discovery of novel SMs. Author summary: The discovery and description of novel fungal secondary metabolites promises novel antibiotics, pharmaceuticals, and other useful compounds. A way to identify novel secondary metabolites is to express the corresponding genes in a suitable expression host. Consequently, a detailed knowledge or an accurate prediction of these genes is necessary. In fungi, the genes are co-localized in so-called biosynthetic gene clusters. Notably, the clusters may also contain genes that are not necessary for the biosynthesis of the secondary metabolites, so-called gap genes. We developed a method to detect co-evolved genes within the clusters and demonstrated that essential genes are co-evolving and can thus be differentiated from the gap genes. This adds an additional layer of information, which can support researchers with their decisions on which genes to study and express for the discovery of novel secondary metabolites. [ABSTRACT FROM AUTHOR]

Rationally and efficiently modifying the amino-acid sequence of proteins to control their ability to undergo liquid–liquid phase separation (LLPS) on demand is not only highly desirable, but can also help to elucidate which protein features are important for LLPS. Here, we propose a computational method that couples a genetic algorithm to a sequence-dependent coarse-grained protein model to evolve the amino-acid sequences of phase-separating intrinsically disordered protein regions (IDRs), and purposely enhance or inhibit their capacity to phase-separate. We validate the predicted critical solution temperatures of the mutated sequences with ABSINTH, a more accurate all-atom model. We apply the algorithm to the phase-separating IDRs of three naturally occurring proteins, namely FUS, hnRNPA1 and LAF1, as prototypes of regions that exist in cells and undergo homotypic LLPS driven by different types of intermolecular interaction, and we find that the evolution of amino-acid sequences towards enhanced LLPS is driven in these three cases, among other factors, by an increase in the average size of the amino acids. However, the direction of change in the molecular driving forces that enhance LLPS (such as hydrophobicity, aromaticity and charge) depends on the initial amino-acid sequence. Finally, we show that the evolution of amino-acid sequences to modulate LLPS is strongly coupled to the make-up of the medium (e.g. the presence or absence of RNA), which may have significant implications for our understanding of phase separation within the many-component mixtures of biological systems. Author summary: Protein condensates formed by the process of liquid–liquid phase separation (LLPS) play diverse roles inside cells—from spatio-temporal compartmentalisation to speeding up chemical reactions. When things go wrong, LLPS can have pathological implications. This realisation has boosted the interest in devising approaches to design rationally amino-acid sequence variations to modulate or even reverse the phase behaviour of proteins on demand. Here, we develop an efficient computational method that combines a genetic algorithm with a sequence-dependent coarse-grained model, and an all-atom model for validation, to identify amino-acid sequence variations of intrinsically disordered proteins that intentionally promote or inhibit their LLPS. Our method can be applied to proteins in pure form and within multi-component systems. [ABSTRACT FROM AUTHOR]

Ankyrin protein repeats bind to a wide range of substrates and are one of the most common protein motifs in nature. Here, we collate a high-quality alignment of 7,407 ankyrin repeats and examine for the first time, the distribution of human population variants from large-scale sequencing of healthy individuals across this family. Population variants are not randomly distributed across the genome but are constrained by gene essentiality and function. Accordingly, we interpret the population variants in context with evolutionary constraint and structural features including secondary structure, accessibility and protein-protein interactions across 383 three-dimensional structures of ankyrin repeats. We find five positions that are highly conserved across homologues and also depleted in missense variants within the human population. These positions are significantly enriched in intra-domain contacts and so likely to be key for repeat packing. In contrast, a group of evolutionarily divergent positions are found to be depleted in missense variants in human but significantly enriched in protein-protein interactions. Our analysis also suggests the domain has three, not two surfaces, each with different patterns of enrichment in protein-substrate interactions and missense variants. Our findings will be of interest to those studying or engineering ankyrin-repeat containing proteins as well as those interpreting the significance of disease variants. Author summary: Comparison of variation at each position of the amino acid sequence for a protein across different species is a powerful way to identify parts of the protein that are important for its structure and function. Large-scale DNA sequencing of healthy people has recently made it possible to study normal genetic variation within just one species. Our work combines information on genetic differences between over 100,000 people with in-depth analysis of all available three-dimensional structures for Ankyrin repeats, which are a widespread family of binding proteins formed by units with similar amino acid sequence that are found in tandem. Our combined analysis identifies sites critical for ankyrin stability as well as the positions most important for substrate interactions and hence function. Although focused only on the Ankyrins, the principles developed in our work are general and can be applied to any protein family. [ABSTRACT FROM AUTHOR]