Search

Protein generation has numerous applications in designing therapeutic antibodies and creating new drugs. Still, it is a demanding task due to the inherent complexities of protein structures and the limitations of current generative models. Proteins possess intricate geometry, and sampling their conformational space is challenging due to its high dimensionality. This paper introduces novel Markovian and non-Markovian generative diffusion models based on fractional stochastic differential equations and the Lévy distribution, allowing for a more effective exploration of the conformational space. The approach is applied to a dataset of 40,000 proteins and evaluated in terms of Fréchet distance, fidelity, and diversity, outperforming the state-of-the-art by 25.4%, 35.8%, and 11.8%, respectively.

This paper presents a new approach for protein generation based on one-shot learning and hybrid quantum neural networks. Given a single protein complex, the system learns how to predict the remaining unknown properties, without resorting to autoregression, from the physicochemical properties of the receptor and a prior on the physicochemical properties of the ligand. In contrast with other approaches, QuantumBound learns from a single instance, not from a large dataset, as is common in deep learning. By knowing half of the properties of the ligand, the system can predict the remaining half with an average relative error of 1.43% for a dataset consisting of one hundred and twenty Covid-19 spikes complexes. To the best of our knowledge, this is the first time that one-shot learning and hybrid quantum computing have been applied to protein generation.

Abstract The design of binder proteins for specific target proteins using deep learning is a challenging task that has a wide range of applications in both designing therapeutic antibodies and creating new drugs. Machine learning-based solutions, as opposed to laboratory design, streamline the design process and enable the design of new proteins that may be required to address new and orphan diseases. Most techniques proposed in the literature necessitate either domain knowledge or some appraisal of the target protein’s 3-D structure. This paper proposes an approach for designing binder proteins based solely on the amino acid sequence of the target protein and without recourse to domain knowledge or structural information. The sequences of the binders are generated with two new transformers, namely the AppendFormer and MergeFormer architectures. Because, in general, there is more than one binder for a given target protein, these transformers employ a binding score and a prior on the sequence of the binder to obtain a unique targeted solution. Our experimental evaluation confirms the strengths of this novel approach. The performance of the models was determined with 5-fold cross-validation and clearly indicates that our architectures lead to highly accurate results. In addition, scores of up to 0.98 were achieved in terms of Needleman-Wunsch and Smith-Waterman similarity metrics, which indicates that our solutions significantly outperform a seq2seq baseline model.

Proteins mainly perform their functions by interacting with other proteins. Protein–protein interactions underpin various biological activities such as metabolic cycles, signal transduction, and immune response. However, due to the sheer number of proteins, experimental methods for finding interacting and non-interacting protein pairs are time-consuming and costly. We therefore developed the ProtInteract framework to predict protein–protein interaction. ProtInteract comprises two components: first, a novel autoencoder architecture that encodes each protein’s primary structure to a lower-dimensional vector while preserving its underlying sequence attributes. This leads to faster training of the second network, a deep convolutional neural network (CNN) that receives encoded proteins and predicts their interaction under three different scenarios. In each scenario, the deep CNN predicts the class of a given encoded protein pair. Each class indicates different ranges of confidence scores corresponding to the probability of whether a predicted interaction occurs or not. The proposed framework features significantly low computational complexity and relatively fast response. The contributions of this work are twofold. First, ProtInteract assimilates the protein’s primary structure into a pseudo-time series. Therefore, we leverage the nature of the time series of proteins and their physicochemical properties to encode a protein’s amino acid sequence into a lower-dimensional vector space. This approach enables extracting highly informative sequence attributes while reducing computational complexity. Second, the ProtInteract framework utilises this information to identify protein interactions with other proteins based on its amino acid configuration. Our results suggest that the proposed framework performs with high accuracy and efficiency in predicting protein-protein interactions.

Most proteins perform their biological function by interacting with themselves or other molecules. Thus, one may obtain biological insights into protein functions, disease prevalence, and therapy development by identifying protein–protein interactions (PPI). However, finding the interacting and non-interacting protein pairs through experimental approaches is labour-intensive and time-consuming, owing to the variety of proteins. Hence, protein–protein interaction and protein–ligand binding problems have drawn attention in the fields of bioinformatics and computer-aided drug discovery. Deep learning methods paved the way for scientists to predict the 3-D structure of proteins from genomes, predict the functions and attributes of a protein, and modify and design new proteins to provide desired functions. This review focuses on recent deep learning methods applied to problems including predicting protein functions, protein–protein interaction and their sites, protein–ligand binding, and protein design.

This work introduces novel approaches, based on geometrical deep learning, for predicting protein–protein interactions. A dataset containing both interacting and non-interacting proteins is selected from the Negatome Database. Interactions are predicted from a graph representing the proteins’ three-dimensional macromolecular surfaces. The nodes are described with heat and wave kernel signatures. Twenty-one neural network architectures are proposed and compared; these are based on graph convolutional neural networks, spectral convolutional neural networks, and a novel spatio–spectral spatialized-gated convolutional neural network. The experimental results demonstrate the accuracy and the efficiency of the proposed architectures.

This paper presents a new deep learning framework, QuantumPath, for long-term stock price prediction, which is of great significance in portfolio management and risk mitigation, especially when the market becomes volatile due to unpredictable circumstances such as a pandemic. Our approach is based on stochastic equations, the Feynman–Dirac path integral, deep Bayesian networks, and temporal generative adversarial neural networks (t-GAN). The expected financial trajectory is evaluated with a Feynman–Dirac path integral. The latter involves summing all possible financial trajectories that could have been taken by the financial instrument. These trajectories are generated with a t-GAN. A probability is attributed to each point of each path. The probability is a function of the Lagrangian, which is derived from a stochastic equation describing the temporal evolution of the stock. The drift and the volatility at each point, which are required in order to evaluate the Lagrangian, are predicted with a deep Bayesian neural network. Given that the evolution of a stock’s price is isomorphic to a time series, our temporal GAN consists of long short-term memory (LSTM) neural networks, which introduce a memory mechanism, and temporal convolutional neural networks (TCN), which ensure causality. Stock prices are predicted over periods of twenty and thirty days for nine stocks, eight of which are included in the S&P 500 index. Our experimental results clearly demonstrate the efficiency of our approach.

Vaccination is the most effective course of action to prevent influenza. About 150 million doses of influenza vaccines were distributed for the 2015-2016 season in the USA alone according to the Centers for Disease Control and Prevention. Vaccine dosage is calculated based on the concentration of hemagglutinin (HA), the main surface glycoprotein expressed by influenza which varies from strain to strain. Therefore yearly-updated strain-specific antibodies and calibrating antigens are required. Preparing these quantification reagents can take up to three months and significantly slows down the release of new vaccine lots. Therefore, to circumvent the need for strain-specific sera, two anti-HA monoclonal antibodies (mAbs) against a highly conserved sequence have been produced by immunizing mice with a novel peptide-conjugate. Immunoblots demonstrate that 40 strains of influenza encompassing HA subtypes H1 to H13, as well as B strains from the Yamagata and Victoria lineage were detected when the two mAbs are combined to from a pan-HA mAb cocktail. Quantification using this pan-HA mAbs cocktail was achieved in a dot blot assay and results correlated with concentrations measured in a hemagglutination assay with a coefficient of correlation of 0.80. A competitive ELISA was also optimised with purified viral-like particles. Regardless of the quantification method used, pan-HA antibodies can be employed to accelerate process development when strain-specific antibodies are not available, and represent a valuable tool in case of pandemics. These antibodies were also expressed in CHO cells to facilitate large-scale production using bioreactor technologies which might be required to meet industrial needs for quantification reagents. Finally, a simulation model was created to predict the binding affinity of the two anti-HA antibodies to the amino acids composing the highly conserved epitope; different probabilities of interaction between a given amino acid and the antibodies might explain the affinity of each antibody against different influenza strains.

Upstream Binding Factor (UBF) is a unique multi-HMGB-box protein first identified as a co-factor in RNA polymerase I (RPI/PolI) transcription. However, its poor DNA sequence selectivity and its ability to generate nucleosome-like nucleoprotein complexes suggest a more generalized role in chromatin structure. We previously showed that extensive depletion of UBF reduced the number of actively transcribed ribosomal RNA (rRNA) genes, but had little effect on rRNA synthesis rates or cell proliferation, leaving open the question of its requirement for RPI transcription. Using gene deletion in mouse, we now show that UBF is essential for embryo development beyond morula. Conditional deletion in cell cultures reveals that UBF is also essential for transcription of the rRNA genes and that it defines the active chromatin conformation of both gene and enhancer sequences. Loss of UBF prevents formation of the SL1/TIF1B pre-initiation complex and recruitment of the RPI-Rrn3/TIF1A complex. It is also accompanied by recruitment of H3K9me3, canonical histone H1 and HP1α, but not by de novo DNA methylation. Further, genes retain penta-acetyl H4 and H2A.Z, suggesting that even in the absence of UBF the rRNA genes can maintain a potentially active state. In contrast to canonical histone H1, binding of H1.4 is dependent on UBF, strongly suggesting that it plays a positive role in gene activity. Unexpectedly, arrest of rRNA synthesis does not suppress transcription of the 5S, tRNA or snRNA genes, nor expression of the several hundred mRNA genes implicated in ribosome biogenesis. Thus, rRNA gene activity does not coordinate global gene expression for ribosome biogenesis. Loss of UBF also unexpectedly induced the formation in cells of a large sub-nuclear structure resembling the nucleolar precursor body (NPB) of oocytes and early embryos. These somatic NPBs contain rRNA synthesis and processing factors but do not associate with the rRNA gene loci (NORs).

This paper presents a new theoretical approach for the description of multidimensional objects for which 3D and 4D are particular cases. The approach is based on a curved space which is associated to each object. This curved space is characterised by Riemannian tensors from which invariant quantities are defined. A descriptor or index is constructed from those invariants for which statistical and abstract graph representations are associated. The obtained representations are invariant under general coordinate transformations. The statistical representation allows a compact description of the object while the abstract graph allows describing the relations in between the parts as well as the structure.

ABSTRACT: The objective of this study was to examine the influence of different environmental factors on ATP luminometry measurements of feeding equipment and to investigate associations with health of preweaning calves and the levels of ATP identified through luminometry. On 50 commercial dairy farms in Quebec, Canada, ATP luminometry measurements (in relative light units, RLU) were obtained using the direct swabbing technique with Hygiena UltraSnap swabs and a liquid rinsing technique with the same swab for automatic milk feeders (AMF), bottles, buckets, esophageal tube feeders (ET), milk replacer, nipples, and water. During this visit, environmental factors (including temperature, air draft, humidity, ammonia, and bacterial count) were collected, and a clinical examination (including respiratory score and fecal score) was performed for all preweaning calves present at the farm. This process was repeated 4 times in a year, leading to collection of luminometer results, environmental parameters, and overall health of calves for each season per farm. Overall, a difference in luminometer results was seen between the different periods sampled for all feeding equipment (except the ET), milk replacer, and water, showing higher RLU values in spring and summer and lower values in autumn and winter. When comparing RLU measurements with environmental factors, only a low to negligible correlation could be found. When feeding equipment was classified as not contaminated or contaminated based on previously described cutoff values, a good agreement within a farm for the different seasons was noticed only for nipples (Gwet's agreement AC1 = 0.64), with a poor to moderate agreement for other feeding equipment. Regarding the different models of nipples, Peach Teat nipples showed higher RLU values compared with Merrick's nipples. An association was seen between the proportion of preweaning calves suffering from diarrhea on the farm and the contamination of AMF based on ATP luminometry (logistic regression estimate = 0.52). For other feeding equipment, milk replacer, and water, no significant associations were found. This study showed that ATP luminometry measurements of feeding equipment from preweaning calves are susceptible to seasonality and type of nipple. Thus, these factors should be taken into consideration when interpreting results. Additionally, an association could be made between diarrhea in preweaning calves and the contamination of AMF based on ATP luminometry, showing the potential clinical importance of this on-farm hygiene assessment tool.