Search

Recurrent neural networks with memory and attention mechanisms are widely used in natural language processing because they can capture short and long term sequential information for diverse tasks. We propose an integrated deep learning model for microbial DNA sequence data, which exploits convolutional neural networks, recurrent neural networks, and attention mechanisms to predict taxonomic classifications and sample-associated attributes, such as the relationship between the microbiome and host phenotype, on the read/sequence level. In this paper, we develop this novel deep learning approach and evaluate its application to amplicon sequences. We apply our approach to short DNA reads and full sequences of 16S ribosomal RNA (rRNA) marker genes, which identify the heterogeneity of a microbial community sample. We demonstrate that our implementation of a novel attention-based deep network architecture, Read2Pheno, achieves read-level phenotypic prediction. Training Read2Pheno models will encode sequences (reads) into dense, meaningful representations: learned embedded vectors output from the intermediate layer of the network model, which can provide biological insight when visualized. The attention layer of Read2Pheno models can also automatically identify nucleotide regions in reads/sequences which are particularly informative for classification. As such, this novel approach can avoid pre/post-processing and manual interpretation required with conventional approaches to microbiome sequence classification. We further show, as proof-of-concept, that aggregating read-level information can robustly predict microbial community properties, host phenotype, and taxonomic classification, with performance at least comparable to conventional approaches. An implementation of the attention-based deep learning network is available at https://github.com/EESI/sequence_attention (a python package) and https://github.com/EESI/seq2att (a command line tool)., Author summary Microbiomes are communities of microscopic organisms found everywhere, including on and in the human body. For example, the gut microbiome plays an important role in digestion, and changes in composition are associated with changes in health or disease, e.g., inflammatory bowel disease (IBD). Today, microbiome composition is often obtained from high-throughput sequencing, which generates many short DNA reads from multiple organisms in a sample. In this paper, we present a novel deep learning framework, Read2Pheno, to predict phenotype from all the reads in a set of biological samples. An attention mechanism allows visualization of specific subregions (sets of bases) which are important in classifying the reads according to phenotype or taxon labels. We evaluate the framework on sequencing data for 16S rRNA genes, genetic markers used to identify microbial taxonomy. We show that Read2Pheno performs comparably as conventional methods on three distinct data sets from the American Gut Project, IBD patients and controls, and a comprehensive taxonomic database. Moreover, Read2Pheno results can be readily interpreted—e.g., to identify regions of the 16S rRNA gene to target for PCR diagnostics—without additional pre/post-processing steps that can introduce complexity and error.

LncRNA-protein interactions play important roles in post-transcriptional gene regulation, poly-adenylation, splicing and translation. Identification of lncRNA-protein interactions helps to understand lncRNA-related activities. Existing computational methods utilize multiple lncRNA features or multiple protein features to predict lncRNA-protein interactions, but features are not available for all lncRNAs or proteins; most of existing methods are not capable of predicting interacting proteins (or lncRNAs) for new lncRNAs (or proteins), which don’t have known interactions. In this paper, we propose the sequence-based feature projection ensemble learning method, “SFPEL-LPI”, to predict lncRNA-protein interactions. First, SFPEL-LPI extracts lncRNA sequence-based features and protein sequence-based features. Second, SFPEL-LPI calculates multiple lncRNA-lncRNA similarities and protein-protein similarities by using lncRNA sequences, protein sequences and known lncRNA-protein interactions. Then, SFPEL-LPI combines multiple similarities and multiple features with a feature projection ensemble learning frame. In computational experiments, SFPEL-LPI accurately predicts lncRNA-protein associations and outperforms other state-of-the-art methods. More importantly, SFPEL-LPI can be applied to new lncRNAs (or proteins). The case studies demonstrate that our method can find out novel lncRNA-protein interactions, which are confirmed by literature. Finally, we construct a user-friendly web server, available at http://www.bioinfotech.cn/SFPEL-LPI/., Author summary LncRNA-protein interactions play important roles in post-transcriptional gene regulation, poly-adenylation, splicing and translation. Identification of lncRNA-protein interactions helps to understand lncRNA-related activities. In this paper, we propose a novel computational method “SFPEL-LPI” to predict lncRNA-protein interactions. SFPEL-LPI makes use of lncRNA sequences, protein sequences and known lncRNA-protein associations to extract features and calculate similarities for lncRNAs and proteins, and then combines them with a feature projection ensemble learning frame. SFPEL-LPI can predict unobserved interactions between lncRNAs and proteins, and also can make predictions for new lncRNAs (or proteins), which have no interactions with any proteins (or lncRNAs). SFPEL-LPI produces high-accuracy performances on the benchmark dataset when evaluated by five-fold cross validation, and outperforms state-of-the-art methods. The case studies demonstrate that SFPEL-LPI can find out novel associations, which are confirmed by literature. To facilitate the lncRNA-protein interaction prediction, we develop a user-friendly web server, available at http://www.bioinfotech.cn/SFPEL-LPI/.

Metaproteomic studies adopt the common bottom-up proteomics approach to investigate the protein composition and the dynamics of protein expression in microbial communities. When matched metagenomic and/or metatranscriptomic data of the microbial communities are available, metaproteomic data analyses often employ a metagenome-guided approach, in which complete or fragmental protein-coding genes are first directly predicted from metagenomic (and/or metatranscriptomic) sequences or from their assemblies, and the resulting protein sequences are then used as the reference database for peptide/protein identification from MS/MS spectra. This approach is often limited because protein coding genes predicted from metagenomes are incomplete and fragmental. In this paper, we present a graph-centric approach to improving metagenome-guided peptide and protein identification in metaproteomics. Our method exploits the de Bruijn graph structure reported by metagenome assembly algorithms to generate a comprehensive database of protein sequences encoded in the community. We tested our method using several public metaproteomic datasets with matched metagenomic and metatranscriptomic sequencing data acquired from complex microbial communities in a biological wastewater treatment plant. The results showed that many more peptides and proteins can be identified when assembly graphs were utilized, improving the characterization of the proteins expressed in the microbial communities. The additional proteins we identified contribute to the characterization of important pathways such as those involved in degradation of chemical hazards. Our tools are released as open-source software on github at https://github.com/COL-IU/Graph2Pro., Author Summary In recent years, meta-omic (including metatranscriptomic and metaproteomic) techniques have been adopted as complementary approaches to metagenomic sequencing to study functional characteristics and dynamics of microbial communities, aiming at a holistic understanding of a community to respond to the changes in the environment. Currently, metaproteomic data are largely analyzed using the bioinformatics tools originally designed in bottom-up proteomics. In particular, recent metaproteomic studies employed a metagenome-guided approach, in which complete or fragmental protein-coding genes were first predicted from metagenomic sequences (i.e., contigs or scaffolds), acquired from the matched community samples, and predicted protein sequences were then used in peptide identification. A key challenge of this approach is that the protein coding genes predicted from assembled metagenomic contigs can be incomplete and fragmented due to the complexity of metagenomic samples and the short reads length in metagenomic sequencing. To address this issue, in this paper, we present a graph-centric approach that exploits the de bruijn graph structure reported by metagenome assembly algorithms to improve metagenome-guided peptide and protein identification in metaproteomics. We show that our method can identify much more peptides and proteins, improving the characterization of the proteins expressed in the microbial communities.

There has been a spate of interest in association networks in biological and medical research, for example, genetic interaction networks. In this paper, we propose a novel method, the extended joint hub graphical lasso (EDOHA), to estimate multiple related interaction networks for high dimensional omics data across multiple distinct classes. To be specific, we construct a convex penalized log likelihood optimization problem and solve it with an alternating direction method of multipliers (ADMM) algorithm. The proposed method can also be adapted to estimate interaction networks for high dimensional compositional data such as microbial interaction networks. The performance of the proposed method in the simulated studies shows that EDOHA has remarkable advantages in recognizing class-specific hubs than the existing comparable methods. We also present three applications of real datasets. Biological interpretations of our results confirm those of previous studies and offer a more comprehensive understanding of the underlying mechanism in disease., Author summary Reconstruction of multiple association networks from high dimensional omics data is an important topic, especially in biology. Previous studies focused on estimating different networks and detecting common hubs among all classes. Integration of information over different classes of data while allowing the difference in the hub nodes is also biologically plausible. Therefore, we propose a method, EDOHA, to jointly construct multiple interaction networks with the capacity in finding different hub networks for each class of data. Simulation studies show better performance over conventional methods. The method has been demonstrated in three real-world data.

A current strategy for obtaining haplotype information from several individuals involves short-read sequencing of pooled amplicons, where fragments from each individual is identified by a unique DNA barcode. In this paper, we report a new method to recover the phylogeny of haplotypes from short-read sequences obtained using pooled amplicons from a mixture of individuals, without barcoding. The method, AFPhyloMix, accepts an alignment of the mixture of reads against a reference sequence, obtains the single-nucleotide-polymorphisms (SNP) patterns along the alignment, and constructs the phylogenetic tree according to the SNP patterns. AFPhyloMix adopts a Bayesian inference model to estimate the phylogeny of the haplotypes and their relative abundances, given that the number of haplotypes is known. In our simulations, AFPhyloMix achieved at least 80% accuracy at recovering the phylogenies and relative abundances of the constituent haplotypes, for mixtures with up to 15 haplotypes. AFPhyloMix also worked well on a real data set of kangaroo mitochondrial DNA sequences., Author summary In evolutionary studies, it is customary to obtain homologous sequences from different individuals in a population or a species to construct a phylogeny. Frequently, sequences from different individuals will be identical; we refer to a set of identical sequences as a haplotype. If short-read sequencing technologies are used to obtain sequences from many individuals, the sequence from each individual is tagged with a unique barcode, and a mixed sample of tagged sequences is subsequently sequenced. The tagged sequences can be identified using the appropriate bioinformatics tools, for further downstream analyses. We have developed a novel method, AFPhyloMix, to reconstruct the phylogeny of a mixed sample of homologous sequences, and the relative abundance of different haplotypes, from different individuals without the need for barcoding. AFPhyloMix aligns the short reads obtained to a reference alignment, and identifies the variable sites along the alignment. On the basis of the patterns of nucleotide frequencies at these and neighbouring sites, AFPhyloMix uses a Bayesian inference model to compute the phylogenetic tree and the haplotype relative abundances. Our results show that AFPhyloMix works well on both the simulated data set and the real data set.

Finding non-standard or new metabolic pathways has important applications in metabolic engineering, synthetic biology and the analysis and reconstruction of metabolic networks. Branched metabolic pathways dominate in metabolic networks and depict a more comprehensive picture of metabolism compared to linear pathways. Although progress has been developed to find branched metabolic pathways, few efforts have been made in identifying branched metabolic pathways via atom group tracking. In this paper, we present a pathfinding method called BPFinder for finding branched metabolic pathways by atom group tracking, which aims to guide the synthetic design of metabolic pathways. BPFinder enumerates linear metabolic pathways by tracking the movements of atom groups in metabolic network and merges the linear atom group conserving pathways into branched pathways. Two merging rules based on the structure of conserved atom groups are proposed to accurately merge the branched compounds of linear pathways to identify branched pathways. Furthermore, the integrated information of compound similarity, thermodynamic feasibility and conserved atom groups is also used to rank the pathfinding results for feasible branched pathways. Experimental results show that BPFinder is more capable of recovering known branched metabolic pathways as compared to other existing methods, and is able to return biologically relevant branched pathways and discover alternative branched pathways of biochemical interest. The online server of BPFinder is available at http://114.215.129.245:8080/atomic/. The program, source code and data can be downloaded from https://github.com/hyr0771/BPFinder., Author summary Computational search of branched metabolic pathways is a fundamental problem in metabolic engineering and metabolic network analysis, which provides a systematic way of understanding the metabolism and discovering alternative pathways for synthesis of useful biomolecules. We propose BPFinder, a novel computational approach to identify branched metabolic pathways via atom group tracking. Different from other pathfinding methods using atom tracking, BPFinder tracks the movement of atom groups in metabolic network to find linear atom group conserving pathways, and merge the found linear pathways by the selected branched compounds to generate branched pathways. Based on the structure of conserved atom groups in branched compounds, we design two merging rules for branched compounds: overlapping rule and non-overlapping rule. The user can flexibly adopt these rules to accurately find the branched pathways that contain overlapping/non-overlapping conserved atom groups. BPFinder also enables the user to combine the information of compound similarity, Gibbs free energy of reactions, and conserved atom groups to sort resulting pathways. Compared with other existing methods, BPFinder can more accurately recover the known branched pathways. The alternative branched pathways returned by BPFinder reveal that the user can flexibly utilize our proposed merging rules to discover biochemically meaningful pathways of interest.

Motivation Protein contacts contain key information for the understanding of protein structure and function and thus, contact prediction from sequence is an important problem. Recently exciting progress has been made on this problem, but the predicted contacts for proteins without many sequence homologs is still of low quality and not very useful for de novo structure prediction. Method This paper presents a new deep learning method that predicts contacts by integrating both evolutionary coupling (EC) and sequence conservation information through an ultra-deep neural network formed by two deep residual neural networks. The first residual network conducts a series of 1-dimensional convolutional transformation of sequential features; the second residual network conducts a series of 2-dimensional convolutional transformation of pairwise information including output of the first residual network, EC information and pairwise potential. By using very deep residual networks, we can accurately model contact occurrence patterns and complex sequence-structure relationship and thus, obtain higher-quality contact prediction regardless of how many sequence homologs are available for proteins in question. Results Our method greatly outperforms existing methods and leads to much more accurate contact-assisted folding. Tested on 105 CASP11 targets, 76 past CAMEO hard targets, and 398 membrane proteins, the average top L long-range prediction accuracy obtained by our method, one representative EC method CCMpred and the CASP11 winner MetaPSICOV is 0.47, 0.21 and 0.30, respectively; the average top L/10 long-range accuracy of our method, CCMpred and MetaPSICOV is 0.77, 0.47 and 0.59, respectively. Ab initio folding using our predicted contacts as restraints but without any force fields can yield correct folds (i.e., TMscore>0.6) for 203 of the 579 test proteins, while that using MetaPSICOV- and CCMpred-predicted contacts can do so for only 79 and 62 of them, respectively. Our contact-assisted models also have much better quality than template-based models especially for membrane proteins. The 3D models built from our contact prediction have TMscore>0.5 for 208 of the 398 membrane proteins, while those from homology modeling have TMscore>0.5 for only 10 of them. Further, even if trained mostly by soluble proteins, our deep learning method works very well on membrane proteins. In the recent blind CAMEO benchmark, our fully-automated web server implementing this method successfully folded 6 targets with a new fold and only 0.3L-2.3L effective sequence homologs, including one β protein of 182 residues, one α+β protein of 125 residues, one α protein of 140 residues, one α protein of 217 residues, one α/β of 260 residues and one α protein of 462 residues. Our method also achieved the highest F1 score on free-modeling targets in the latest CASP (Critical Assessment of Structure Prediction), although it was not fully implemented back then. Availability http://raptorx.uchicago.edu/ContactMap/, Author Summary Protein contact prediction and contact-assisted folding has made good progress due to direct evolutionary coupling analysis (DCA). However, DCA is effective on only some proteins with a very large number of sequence homologs. To further improve contact prediction, we borrow ideas from deep learning, which has recently revolutionized object recognition, speech recognition and the GO game. Our deep learning method can model complex sequence-structure relationship and high-order correlation (i.e., contact occurrence patterns) and thus, improve contact prediction accuracy greatly. Our test results show that our method greatly outperforms the state-of-the-art methods regardless how many sequence homologs are available for a protein in question. Ab initio folding guided by our predicted contacts may fold many more test proteins than the other contact predictors. Our contact-assisted 3D models also have much better quality than homology models built from the training proteins, especially for membrane proteins. One interesting finding is that even trained mostly with soluble proteins, our method performs very well on membrane proteins. Recent blind CAMEO test confirms that our method can fold large proteins with a new fold and only a small number of sequence homologs.

The rapid development of sequencing technology has led to an explosive accumulation of genomic sequence data. Clustering is often the first step to perform in sequence analysis, and hierarchical clustering is one of the most commonly used approaches for this purpose. However, it is currently computationally expensive to perform hierarchical clustering of extremely large sequence datasets due to its quadratic time and space complexities. In this paper we developed a new algorithm called ESPRIT-Forest for parallel hierarchical clustering of sequences. The algorithm achieves subquadratic time and space complexity and maintains a high clustering accuracy comparable to the standard method. The basic idea is to organize sequences into a pseudo-metric based partitioning tree for sub-linear time searching of nearest neighbors, and then use a new multiple-pair merging criterion to construct clusters in parallel using multiple threads. The new algorithm was tested on the human microbiome project (HMP) dataset, currently one of the largest published microbial 16S rRNA sequence dataset. Our experiment demonstrated that with the power of parallel computing it is now compu- tationally feasible to perform hierarchical clustering analysis of tens of millions of sequences. The software is available at http://www.acsu.buffalo.edu/∼yijunsun/lab/ESPRIT-Forest.html.

Approaches to identify significant pathways from high-throughput quantitative data have been developed in recent years. Still, the analysis of proteomic data stays difficult because of limited sample size. This limitation also leads to the practice of using a competitive null as common approach; which fundamentally implies genes or proteins as independent units. The independent assumption ignores the associations among biomolecules with similar functions or cellular localization, as well as the interactions among them manifested as changes in expression ratios. Consequently, these methods often underestimate the associations among biomolecules and cause false positives in practice. Some studies incorporate the sample covariance matrix into the calculation to address this issue. However, sample covariance may not be a precise estimation if the sample size is very limited, which is usually the case for the data produced by mass spectrometry. In this study, we introduce a multivariate test under a self-contained null to perform pathway analysis for quantitative proteomic data. The covariance matrix used in the test statistic is constructed by the confidence scores retrieved from the STRING database or the HitPredict database. We also design an integrating procedure to retain pathways of sufficient evidence as a pathway group. The performance of the proposed T2-statistic is demonstrated using five published experimental datasets: the T-cell activation, the cAMP/PKA signaling, the myoblast differentiation, and the effect of dasatinib on the BCR-ABL pathway are proteomic datasets produced by mass spectrometry; and the protective effect of myocilin via the MAPK signaling pathway is a gene expression dataset of limited sample size. Compared with other popular statistics, the proposed T2-statistic yields more accurate descriptions in agreement with the discussion of the original publication. We implemented the T2-statistic into an R package T2GA, which is available at https://github.com/roqe/T2GA., Author summary Pathway analysis is a common approach to quickly access the pathways being regulated in the experiments. There are numerous statistics to perform pathway analysis; most of them assume that the genes or proteins are independent of each other for statistical ease. This assumption, however, is unrealistic to the real biological system and may cause false positives in practice. A standard way to address this issue is to measure the associations among genes or proteins. Unfortunately, the estimation of associations requires sufficient sample size, which is usually not available for proteomic data produced by mass spectrometry. In this study, we propose a T2-statistic, which estimates the associations among gene products, to perform pathway analysis for quantitative proteomic data. Instead of calculating the associations directly from data, we use the confidence scores retrieved from protein-protein interaction databases. We also design an integrating procedure to reserve pathways of sufficient evidence as a regulated pathway group. We compare the proposed T2-statistic to other popular statistics using five published experimental datasets, and the T2-statistic yields more accurate descriptions in agreement with the discussion of the original papers.