Proteins primarily perform their functions through interactions with other proteins, making the accurate prediction of protein-protein interactions (PPIs) a fundamental problem. Experimental methods for determining PPIs are often slow and expensive, which has driven significant efforts to improve the performance of computational methods in this field. While many methods have been designed, recent thorough investigations proved that the existing methods learn exclusively from sequence similarities and node degrees. When such data leakage is avoided, performances were shown to become random. We introduce C3PI, a novel sequence-based deep learning framework designed for predicting PPIs. C3PI uses as input ProtT5 protein embeddings into a complex architecture that includes two novel components, a puzzler and an entangler, which significantly enhance the model's performance. Through extensive comparisons with state-of-the-art methods across many datasets, C3PI consistently outperforms competing approaches, especially in key metrics such as AUPRC and AUROC. Most importantly, C3PI is the first PPI prediction method to achieve a significant improvement over random on the leakage-free gold standard dataset. C3PI is available as a web server at c3pi.csd.uwo.ca and source code from github.com/lucian-ilie/C3PI.

Sequence similarity is of paramount importance in biology, as similar sequences tend to have similar function and share common ancestry. Scoring matrices, such as PAM or BLOSUM, play a crucial role in all bioinformatics algorithms for identifying similarities, but have the drawback that they are fixed, independent of context. We propose a new scoring method for amino acid similarity that remedies this weakness, being contextually dependent. It relies on recent advances in deep learning architectures that employ self-supervised learning in order to leverage the power of enormous amounts of unlabelled data to generate contextual embeddings, which are vector representations for words. These ideas have been applied to protein sequences, producing embedding vectors for protein residues. We propose the E-score between two residues as the cosine similarity between their embedding vector representations. Thorough testing on a wide variety of reference multiple sequence alignments indicate that the alignments produced using the new $E$-score method, especially ProtT5-score, are significantly better than those obtained using BLOSUM matrices. The new method proposes to change the way alignments are computed, with far-reaching implications in all areas of textual data that use sequence similarity. The program to compute alignments based on various $E$-scores is available as a web server at e-score.csd.uwo.ca. The source code is freely available for download from github.com/lucian-ilie/E-score.

MOTIVATION: Proteins accomplish cellular functions by interacting with each other, which makes the prediction of interaction sites a fundamental problem. As experimental methods are expensive and time consuming, computational prediction of the interaction sites has been studied extensively. Structure-based programs are the most accurate, while the sequence-based ones are much more widely applicable, as the sequences available outnumber the structures by two orders of magnitude. Ideally, we would like a tool that has the quality of the former and the applicability of the latter. RESULTS: We provide here the first solution that achieves these two goals. Our new sequence-based program, Seq-InSite, greatly surpasses the performance of sequence-based models, matching the quality of state-of-the-art structure-based predictors, thus effectively superseding the need for models requiring structure. The predictive power of Seq-InSite is illustrated using an analysis of evolutionary conservation for four protein sequences. AVAILABILITY AND IMPLEMENTATION: Seq-InSite is freely available as a web server at http://seq-insite.csd.uwo.ca/ and as free source code, including trained models and all datasets used for training and testing, at https://github.com/lucian-ilie/Seq-InSite.

BACKGROUND: Identification of biomarkers, which are measurable characteristics of biological datasets, can be challenging. Although amplicon sequence variants (ASVs) can be considered potential biomarkers, identifying important ASVs in high-throughput sequencing datasets is challenging. Noise, algorithmic failures to account for specific distributional properties, and feature interactions can complicate the discovery of ASV biomarkers. In addition, these issues can impact the replicability of various models and elevate false-discovery rates. Contemporary machine learning approaches can be leveraged to address these issues. Ensembles of decision trees are particularly effective at classifying the types of data commonly generated in high-throughput sequencing (HTS) studies due to their robustness when the number of features in the training data is orders of magnitude larger than the number of samples. In addition, when combined with appropriate model introspection algorithms, machine learning algorithms can also be used to discover and select potential biomarkers. However, the construction of these models could introduce various biases which potentially obfuscate feature discovery. RESULTS: We developed a decision tree ensemble, LANDMark, which uses oblique and non-linear cuts at each node. In synthetic and toy tests LANDMark consistently ranked as the best classifier and often outperformed the Random Forest classifier. When trained on the full metabarcoding dataset obtained from Canada's Wood Buffalo National Park, LANDMark was able to create highly predictive models and achieved an overall balanced accuracy score of 0.96 ± 0.06. The use of recursive feature elimination did not impact LANDMark's generalization performance and, when trained on data from the BE amplicon, it was able to outperform the Linear Support Vector Machine, Logistic Regression models, and Stochastic Gradient Descent models (p ≤ 0.05). Finally, LANDMark distinguishes itself due to its ability to learn smoother non-linear decision boundaries. CONCLUSIONS: Our work introduces LANDMark, a meta-classifier which blends the characteristics of several machine learning models into a decision tree and ensemble learning framework. To our knowledge, this is the first study to apply this type of ensemble approach to amplicon sequencing data and we have shown that analyzing these datasets using LANDMark can produce highly predictive and consistent models.

MOTIVATION: Proteins usually perform their functions by interacting with other proteins, which is why accurately predicting protein-protein interaction (PPI) binding sites is a fundamental problem. Experimental methods are slow and expensive. Therefore, great efforts are being made towards increasing the performance of computational methods. RESULTS: We propose DEep Learning Prediction of Highly probable protein Interaction sites (DELPHI), a new sequence-based deep learning suite for PPI-binding sites prediction. DELPHI has an ensemble structure which combines a CNN and a RNN component with fine tuning technique. Three novel features, HSP, position information and ProtVec are used in addition to nine existing ones. We comprehensively compare DELPHI to nine state-of-the-art programmes on five datasets, and DELPHI outperforms the competing methods in all metrics even though its training dataset shares the least similarities with the testing datasets. In the most important metrics, AUPRC and MCC, it surpasses the second best programmes by as much as 18.5% and 27.7%, respectively. We also demonstrated that the improvement is essentially due to using the ensemble model and, especially, the three new features. Using DELPHI it is shown that there is a strong correlation with protein-binding residues (PBRs) and sites with strong evolutionary conservation. In addition, DELPHI's predicted PBR sites closely match known data from Pfam. DELPHI is available as open-sourced standalone software and web server. AVAILABILITY AND IMPLEMENTATION: The DELPHI web server can be found at delphi.csd.uwo.ca/, with all datasets and results in this study. The trained models, the DELPHI standalone source code, and the feature computation pipeline are freely available at github.com/lucian-ilie/DELPHI. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

SUMMARY: Continuous-time Markov chain models with finite state space are routinely used for analysis of discrete character data on phylogenetic trees. Examples of such discrete character data include restriction sites, gene family presence/absence, intron presence/absence and gene family size data. While models with constrained substitution rate matrices have been used to good effect, more biologically realistic models have been increasingly implemented in the recent literature combining, e.g., site rate variation, site partitioning, branch-specific rates, allowing for non-stationary prior root probabilities, correcting for sampling bias, etc. to name a few. Here, a flexible and fast R package is introduced that infers evolutionary rates of discrete characters on a tree within a probabilistic framework. The package, markophylo, fits maximum-likelihood models using Markov chains on phylogenetic trees. The package is efficient, with the workhorse functions written in C++ and the interface in user-friendly R. AVAILABILITY AND IMPLEMENTATION: markophylo is available as a platform-independent R package from the Comprehensive R Archive Network at https://cran.r-project.org/web/packages/markophylo/. A vignette with numerous examples is also provided with the R package. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: A number of statistical phylogenetic methods have been developed to infer conserved functional sites or regions in proteins. Many methods, e.g. Rate4Site, apply the standard phylogenetic models to infer site-specific substitution rates and totally ignore the spatial correlation of substitution rates in protein tertiary structures, which may reduce their power to identify conserved functional patches in protein tertiary structures when the sequences used in the analysis are highly similar. The 3D sliding window method has been proposed to infer conserved functional patches in protein tertiary structures, but the window size, which reflects the strength of the spatial correlation, must be predefined and is not inferred from data. We recently developed GP4Rate to solve these problems under the Bayesian framework. Unfortunately, GP4Rate is computationally slow. Here, we present an intuitive web server, FuncPatch, to perform a fast approximate Bayesian inference of conserved functional patches in protein tertiary structures. RESULTS: Both simulations and four case studies based on empirical data suggest that FuncPatch is a good approximation to GP4Rate. However, FuncPatch is orders of magnitudes faster than GP4Rate. In addition, simulations suggest that FuncPatch is potentially a useful tool complementary to Rate4Site, but the 3D sliding window method is less powerful than FuncPatch and Rate4Site. The functional patches predicted by FuncPatch in the four case studies are supported by experimental evidence, which corroborates the usefulness of FuncPatch. AVAILABILITY AND IMPLEMENTATION: The software FuncPatch is freely available at the web site, http://info.mcmaster.ca/yifei/FuncPatch CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

A critical question in biology is the identification of functionally important amino acid sites in proteins. Because functionally important sites are under stronger purifying selection, site-specific substitution rates tend to be lower than usual at these sites. A large number of phylogenetic models have been developed to estimate site-specific substitution rates in proteins and the extraordinarily low substitution rates have been used as evidence of function. Most of the existing tools, e.g. Rate4Site, assume that site-specific substitution rates are independent across sites. However, site-specific substitution rates may be strongly correlated in the protein tertiary structure, since functionally important sites tend to be clustered together to form functional patches. We have developed a new model, GP4Rate, which incorporates the Gaussian process model with the standard phylogenetic model to identify slowly evolved regions in protein tertiary structures. GP4Rate uses the Gaussian process to define a nonparametric prior distribution of site-specific substitution rates, which naturally captures the spatial correlation of substitution rates. Simulations suggest that GP4Rate can potentially estimate site-specific substitution rates with a much higher accuracy than Rate4Site and tends to report slowly evolved regions rather than individual sites. In addition, GP4Rate can estimate the strength of the spatial correlation of substitution rates from the data. By applying GP4Rate to a set of mammalian B7-1 genes, we found a highly conserved region which coincides with experimental evidence. GP4Rate may be a useful tool for the in silico prediction of functionally important regions in the proteins with known structures.

BACKGROUND: DNA microarrays have become ubiquitous in biological and medical research. The most difficult problem that needs to be solved is the design of DNA oligonucleotides that (i) are highly specific, that is, bind only to the intended target, (ii) cover the highest possible number of genes, that is, all genes that allow such unique regions, and (iii) are computed fast. None of the existing programs meet all these criteria. RESULTS: We introduce a new approach with our software program BOND (Basic OligoNucleotide Design). According to Kane's criteria for oligo design, BOND computes highly specific DNA oligonucleotides, for all the genes that admit unique probes, while running orders of magnitude faster than the existing programs. The same approach enables us to introduce also an evaluation procedure that correctly measures the quality of the oligonucleotides. Extensive comparison is performed to prove our claims. BOND is flexible, easy to use, requires no additional software, and is freely available for non-commercial use from http://www.csd.uwo.ca/∼ilie/BOND/. CONCLUSIONS: We provide an improved solution to the important problem of oligonucleotide design, including a thorough evaluation of oligo design programs. We hope BOND will become a useful tool for researchers in biological and medical sciences by making the microarray procedures faster and more accurate.

MOTIVATION: A number of statistical phylogenetic methods have been proposed to identify type-I functional divergence in duplicate genes by detecting heterogeneous substitution rates in phylogenetic trees. A common disadvantage of the existing methods is that autocorrelation of substitution rates along sequences is not modeled. This reduces the power of existing methods to identify regions under functional divergence. RESULTS: We design a phylogenetic hidden Markov model to identify protein regions relevant to type-I functional divergence. A C++ program, HMMDiverge, has been developed to estimate model parameters and to identify regions under type-I functional divergence. Simulations demonstrate that HMMDiverge can successfully identify protein regions under type-I functional divergence unless the discrepancy of substitution rates between subfamilies is very limited or the regions under functional divergence are very short. Applying HMMDiverge to G protein α subunits in animals, we identify a candidate region longer than 20 amino acids, which overlaps with the α-4 helix and the α4-β6 loop in the GTPase domain with divergent rates of substitutions. These sites are different from those reported by an existing program, DIVERGE2. Interestingly, previous biochemical studies suggest the α-4 helix and the α4-β6 loop are important to the specificity of the receptor-G protein interaction. Therefore, the candidate region reported by HMMDiverge highlights that the type-I functional divergence in G protein α subunits may be relevant to the change of receptor-G protein specificity after gene duplication. From these results, we conclude that HMMDiverge is a useful tool to identify regions under type-I functional divergence after gene duplication. AVAILABILITY: C++ source codes of HMMDiverge and simulation programs used in this study, as well as example datasets, are available at http://info.mcmaster.ca/yifei/software/HMMDiverge.html.