Transfer learning improves pMHC kinetic stability and immunogenicity predictions.

The cellular immune response comprises several processes, with the most notable ones being the binding of the peptide to the Major Histocompability Complex (MHC), the peptide-MHC (pMHC) presentation to the surface of the cell, and the recognition of the pMHC by the T-Cell Receptor. Identifying the most potent peptide targets for MHC binding, presentation and T-cell recognition is vital for developing peptide-based vaccines and T-cell-based immunotherapies. Data-driven tools that predict each of these steps have been developed, and the availability of mass spectrometry (MS) datasets has facilitated the development of accurate Machine Learning (ML) methods for class-I pMHC binding prediction. However, the accuracy of ML-based tools for pMHC kinetic stability prediction and peptide immunogenicity prediction is uncertain, as stability and immunogenicity datasets are not abundant. Here, we use transfer learning techniques to improve stability and immunogenicity predictions, by taking advantage of a large number of binding affinity and MS datasets. The resulting models, TLStab and TLImm, exhibit comparable or better performance than state-of-the-art approaches on different stability and immunogenicity test sets respectively. Our approach demonstrates the promise of learning from the task of peptide binding to improve predictions on downstream tasks. The source code of TLStab and TLImm is publicly available at https://github.com/KavrakiLab/TL-MHC.

APE-Gen2.0: Expanding Rapid Class I Peptide-Major Histocompatibility Complex Modeling to Post-Translational Modifications and Noncanonical Peptide Geometries.

The recognition of peptides bound to class I major histocompatibility complex (MHC-I) receptors by T-cell receptors (TCRs) is a determinant of triggering the adaptive immune response. While the exact molecular features that drive the TCR recognition are still unknown, studies have suggested that the geometry of the joint peptide-MHC (pMHC) structure plays an important role. As such, there is a definite need for methods and tools that accurately predict the structure of the peptide bound to the MHC-I receptor. In the past few years, many pMHC structural modeling tools have emerged that provide high-quality modeled structures in the general case. However, there are numerous instances of non-canonical cases in the immunopeptidome that the majority of pMHC modeling tools do not attend to, most notably, peptides that exhibit non-standard amino acids and post-translational modifications (PTMs) or peptides that assume non-canonical geometries in the MHC binding cleft. Such chemical and structural properties have been shown to be present in neoantigens; therefore, accurate structural modeling of these instances can be vital for cancer immunotherapy. To this end, we have developed APE-Gen2.0, a tool that improves upon its predecessor and other pMHC modeling tools, both in terms of modeling accuracy and the available modeling range of non-canonical peptide cases. Some of the improvements include (i) the ability to model peptides that have different types of PTMs such as phosphorylation, nitration, and citrullination; (ii) a new and improved anchor identification routine in order to identify and model peptides that exhibit a non-canonical anchor conformation; and (iii) a web server that provides a platform for easy and accessible pMHC modeling. We further show that structures predicted by APE-Gen2.0 can be used to assess the effects that PTMs have in binding affinity in a more accurate manner than just using solely the sequence of the peptide. APE-Gen2.0 is freely available at https://apegen.kavrakilab.org.

HLAEquity: Examining biases in pan-allele peptide-HLA binding predictors.

Peptide-HLA (pHLA) binding prediction is essential in screening peptide candidates for personalized peptide vaccines. Machine learning (ML) pHLA binding prediction tools are trained on vast amounts of data and are effective in screening peptide candidates. Most ML models report the ability to generalize to HLA alleles unseen during training ("pan-allele" models). However, the use of datasets with imbalanced allele content raises concerns about biased model performance. First, we examine the data bias of two ML-based pan-allele pHLA binding predictors. We find that the pHLA datasets overrepresent alleles from geographic populations of high-income countries. Second, we show that the identified data bias is perpetuated within ML models, leading to algorithmic bias and subpar performance for alleles expressed in low-income geographic populations. We draw attention to the potential therapeutic consequences of this bias, and we challenge the use of the term "pan-allele" to describe models trained with currently available public datasets.

EnGens: a computational framework for generation and analysis of representative protein conformational ensembles.

Proteins are dynamic macromolecules that perform vital functions in cells. A protein structure determines its function, but this structure is not static, as proteins change their conformation to achieve various functions. Understanding the conformational landscapes of proteins is essential to understand their mechanism of action. Sets of carefully chosen conformations can summarize such complex landscapes and provide better insights into protein function than single conformations. We refer to these sets as representative conformational ensembles. Recent advances in computational methods have led to an increase in the number of available structural datasets spanning conformational landscapes. However, extracting representative conformational ensembles from such datasets is not an easy task and many methods have been developed to tackle it. Our new approach, EnGens (short for ensemble generation), collects these methods into a unified framework for generating and analyzing representative protein conformational ensembles. In this work, we: (1) provide an overview of existing methods and tools for representative protein structural ensemble generation and analysis; (2) unify existing approaches in an open-source Python package, and a portable Docker image, providing interactive visualizations within a Jupyter Notebook pipeline; (3) test our pipeline on a few canonical examples from the literature. Representative ensembles produced by EnGens can be used for many downstream tasks such as protein-ligand ensemble docking, Markov state modeling of protein dynamics and analysis of the effect of single-point mutations.

An end-to-end deep learning framework for translating mass spectra to de-novo molecules.

Elucidating the structure of a chemical compound is a fundamental task in chemistry with applications in multiple domains including drug discovery, precision medicine, and biomarker discovery. The common practice for elucidating the structure of a compound is to obtain a mass spectrum and subsequently retrieve its structure from spectral databases. However, these methods fail for novel molecules that are not present in the reference database. We propose Spec2Mol, a deep learning architecture for molecular structure recommendation given mass spectra alone. Spec2Mol is inspired by the Speech2Text deep learning architectures for translating audio signals into text. Our approach is based on an encoder-decoder architecture. The encoder learns the spectra embeddings, while the decoder, pre-trained on a massive dataset of chemical structures for translating between different molecular representations, reconstructs SMILES sequences of the recommended chemical structures. We have evaluated Spec2Mol by assessing the molecular similarity between the recommended structures and the original structure. Our analysis showed that Spec2Mol is able to identify the presence of key molecular substructures from its mass spectrum, and shows on par performance, when compared to existing fragmentation tree methods particularly when test structure information is not available during training or present in the reference database.

EnGens: a computational framework for generation and analysis of representative protein conformational ensembles.

Proteins are dynamic macromolecules that perform vital functions in cells. A protein structure determines its function, but this structure is not static, as proteins change their conformation to achieve various functions. Understanding the conformational landscapes of proteins is essential to understand their mechanism of action. Sets of carefully chosen conformations can summarize such complex landscapes and provide better insights into protein function than single conformations. We refer to these sets as representative conformational ensembles. Recent advances in computational methods have led to an increase in number of available structural datasets spanning conformational landscapes. However, extracting representative conformational ensembles from such datasets is not an easy task and many methods have been developed to tackle it. Our new approach, EnGens (short for ensemble generation), collects these methods into a unified framework for generating and analyzing protein conformational ensembles. In this work we: (1) provide an overview of existing methods and tools for protein structural ensemble generation and analysis; (2) unify existing approaches in an open-source Python package, and a portable Docker image, providing interactive visualizations within a Jupyter Notebook pipeline; (3) test our pipeline on a few canonical examples found in the literature. Representative ensembles produced by EnGens can be used for many downstream tasks such as protein-ligand ensemble docking, Markov state modeling of protein dynamics and analysis of the effect of single-point mutations.

PepSim: T-cell cross-reactivity prediction via comparison of peptide sequence and peptide-HLA structure.

Introduction: Peptide-HLA class I (pHLA) complexes on the surface of tumor cells can be targeted by cytotoxic T-cells to eliminate tumors, and this is one of the bases for T-cell-based immunotherapies. However, there exist cases where therapeutic T-cells directed towards tumor pHLA complexes may also recognize pHLAs from healthy normal cells. The process where the same T-cell clone recognizes more than one pHLA is referred to as T-cell cross-reactivity and this process is driven mainly by features that make pHLAs similar to each other. T-cell cross-reactivity prediction is critical for designing T-cell-based cancer immunotherapies that are both effective and safe. Methods: Here we present PepSim, a novel score to predict T-cell cross-reactivity based on the structural and biochemical similarity of pHLAs. Results and discussion: We show our method can accurately separate cross-reactive from non-crossreactive pHLAs in a diverse set of datasets including cancer, viral, and self-peptides. PepSim can be generalized to work on any dataset of class I peptide-HLAs and is freely available as a web server at pepsim.kavrakilab.org.

Charge-based interactions through peptide position 4 drive diversity of antigen presentation by human leukocyte antigen class I molecules.

Human leukocyte antigen class I (HLA-I) molecules bind and present peptides at the cell surface to facilitate the induction of appropriate CD8+ T cell-mediated immune responses to pathogen- and self-derived proteins. The HLA-I peptide-binding cleft contains dominant anchor sites in the B and F pockets that interact primarily with amino acids at peptide position 2 and the C-terminus, respectively. Nonpocket peptide-HLA interactions also contribute to peptide binding and stability, but these secondary interactions are thought to be unique to individual HLA allotypes or to specific peptide antigens. Here, we show that two positively charged residues located near the top of peptide-binding cleft facilitate interactions with negatively charged residues at position 4 of presented peptides, which occur at elevated frequencies across most HLA-I allotypes. Loss of these interactions was shown to impair HLA-I/peptide binding and complex stability, as demonstrated by both in vitro and in silico experiments. Furthermore, mutation of these Arginine-65 (R65) and/or Lysine-66 (K66) residues in HLA-A*02:01 and A*24:02 significantly reduced HLA-I cell surface expression while also reducing the diversity of the presented peptide repertoire by up to 5-fold. The impact of the R65 mutation demonstrates that nonpocket HLA-I/peptide interactions can constitute anchor motifs that exert an unexpectedly broad influence on HLA-I-mediated antigen presentation. These findings provide fundamental insights into peptide antigen binding that could broadly inform epitope discovery in the context of viral vaccine development and cancer immunotherapy.

SARS-Arena: Sequence and Structure-Guided Selection of Conserved Peptides from SARS-related Coronaviruses for Novel Vaccine Development.

The pandemic caused by the SARS-CoV-2 virus, the agent responsible for the COVID-19 disease, has affected millions of people worldwide. There is constant search for new therapies to either prevent or mitigate the disease. Fortunately, we have observed the successful development of multiple vaccines. Most of them are focused on one viral envelope protein, the spike protein. However, such focused approaches may contribute for the rise of new variants, fueled by the constant selection pressure on envelope proteins, and the widespread dispersion of coronaviruses in nature. Therefore, it is important to examine other proteins, preferentially those that are less susceptible to selection pressure, such as the nucleocapsid (N) protein. Even though the N protein is less accessible to humoral response, peptides from its conserved regions can be presented by class I Human Leukocyte Antigen (HLA) molecules, eliciting an immune response mediated by T-cells. Given the increased number of protein sequences deposited in biological databases daily and the N protein conservation among viral strains, computational methods can be leveraged to discover potential new targets for SARS-CoV-2 and SARS-CoV-related viruses. Here we developed SARS-Arena, a user-friendly computational pipeline that can be used by practitioners of different levels of expertise for novel vaccine development. SARS-Arena combines sequence-based methods and structure-based analyses to (i) perform multiple sequence alignment (MSA) of SARS-CoV-related N protein sequences, (ii) recover candidate peptides of different lengths from conserved protein regions, and (iii) model the 3D structure of the conserved peptides in the context of different HLAs. We present two main Jupyter Notebook workflows that can help in the identification of new T-cell targets against SARS-CoV viruses. In fact, in a cross-reactive case study, our workflows identified a conserved N protein peptide (SPRWYFYYL) recognized by CD8+ T-cells in the context of HLA-B7+. SARS-Arena is available at https://github.com/KavrakiLab/SARS-Arena.

3pHLA-score improves structure-based peptide-HLA binding affinity prediction.

Binding of peptides to Human Leukocyte Antigen (HLA) receptors is a prerequisite for triggering immune response. Estimating peptide-HLA (pHLA) binding is crucial for peptide vaccine target identification and epitope discovery pipelines. Computational methods for binding affinity prediction can accelerate these pipelines. Currently, most of those computational methods rely exclusively on sequence-based data, which leads to inherent limitations. Recent studies have shown that structure-based data can address some of these limitations. In this work we propose a novel machine learning (ML) structure-based protocol to predict binding affinity of peptides to HLA receptors. For that, we engineer the input features for ML models by decoupling energy contributions at different residue positions in peptides, which leads to our novel per-peptide-position protocol. Using Rosetta's ref2015 scoring function as a baseline we use this protocol to develop 3pHLA-score. Our per-peptide-position protocol outperforms the standard training protocol and leads to an increase from 0.82 to 0.99 of the area under the precision-recall curve. 3pHLA-score outperforms widely used scoring functions (AutoDock4, Vina, Dope, Vinardo, FoldX, GradDock) in a structural virtual screening task. Overall, this work brings structure-based methods one step closer to epitope discovery pipelines and could help advance the development of cancer and viral vaccines.

Graph representation learning for structural proteomics.

The field of structural proteomics, which is focused on studying the structure-function relationship of proteins and protein complexes, is experiencing rapid growth. Since the early 2000s, structural databases such as the Protein Data Bank are storing increasing amounts of protein structural data, in addition to modeled structures becoming increasingly available. This, combined with the recent advances in graph-based machine-learning models, enables the use of protein structural data in predictive models, with the goal of creating tools that will advance our understanding of protein function. Similar to using graph learning tools to molecular graphs, which currently undergo rapid development, there is also an increasing trend in using graph learning approaches on protein structures. In this short review paper, we survey studies that use graph learning techniques on proteins, and examine their successes and shortcomings, while also discussing future directions.

DINC-COVID: A webserver for ensemble docking with flexible SARS-CoV-2 proteins.

An unprecedented research effort has been undertaken in response to the ongoing COVID-19 pandemic. This has included the determination of hundreds of crystallographic structures of SARS-CoV-2 proteins, and numerous virtual screening projects searching large compound libraries for potential drug inhibitors. Unfortunately, these initiatives have had very limited success in producing effective inhibitors against SARS-CoV-2 proteins. A reason might be an often overlooked factor in these computational efforts: receptor flexibility. To address this issue we have implemented a computational tool for ensemble docking with SARS-CoV-2 proteins. We have extracted representative ensembles of protein conformations from the Protein Data Bank and from in silico molecular dynamics simulations. Twelve pre-computed ensembles of SARS-CoV-2 protein conformations have now been made available for ensemble docking via a user-friendly webserver called DINC-COVID (dinc-covid.kavrakilab.org). We have validated DINC-COVID using data on tested inhibitors of two SARS-CoV-2 proteins, obtaining good correlations between docking-derived binding energies and experimentally-determined binding affinities. Some of the best results have been obtained on a dataset of large ligands resolved via room temperature crystallography, and therefore capturing alternative receptor conformations. In addition, we have shown that the ensembles available in DINC-COVID capture different ranges of receptor flexibility, and that this diversity is useful in finding alternative binding modes of ligands. Overall, our work highlights the importance of accounting for receptor flexibility in docking studies, and provides a platform for the identification of new inhibitors against SARS-CoV-2 proteins.

DINC-COVID: A webserver for ensemble docking with flexible SARS-CoV-2 proteins.

MOTIVATION: Recent efforts to computationally identify inhibitors for SARS-CoV-2 proteins have largely ignored the issue of receptor flexibility. We have implemented a computational tool for ensemble docking with the SARS-CoV-2 proteins, including the main protease (Mpro), papain-like protease (PLpro) and RNA-dependent RNA polymerase (RdRp). RESULTS: Ensembles of other SARS-CoV-2 proteins are being prepared and made available through a user-friendly docking interface. Plausible binding modes between conformations of a selected ensemble and an uploaded ligand are generated by DINC, our parallelized meta-docking tool. Binding modes are scored with three scoring functions, and account for the flexibility of both the ligand and receptor. Additional details on our methods are provided in the supplementary material. AVAILABILITY: dinc-covid.kavrakilab.org. SUPPLEMENTARY INFORMATION: Details on methods for ensemble generation and docking are provided as supplementary data online. CONTACT: geancarlo.zanatta@ufc.br , kavraki@rice.edu.

Large-Scale Structure-Based Screening of Potential T Cell Cross-Reactivities Involving Peptide-Targets From BCG Vaccine and SARS-CoV-2.

Although not being the first viral pandemic to affect humankind, we are now for the first time faced with a pandemic caused by a coronavirus. The Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) has been responsible for the COVID-19 pandemic, which caused more than 4.5 million deaths worldwide. Despite unprecedented efforts, with vaccines being developed in a record time, SARS-CoV-2 continues to spread worldwide with new variants arising in different countries. Such persistent spread is in part enabled by public resistance to vaccination in some countries, and limited access to vaccines in other countries. The limited vaccination coverage, the continued risk for resistant variants, and the existence of natural reservoirs for coronaviruses, highlight the importance of developing additional therapeutic strategies against SARS-CoV-2 and other coronaviruses. At the beginning of the pandemic it was suggested that countries with Bacillus Calmette-Guérin (BCG) vaccination programs could be associated with a reduced number and/or severity of COVID-19 cases. Preliminary studies have provided evidence for this relationship and further investigation is being conducted in ongoing clinical trials. The protection against SARS-CoV-2 induced by BCG vaccination may be mediated by cross-reactive T cell lymphocytes, which recognize peptides displayed by class I Human Leukocyte Antigens (HLA-I) on the surface of infected cells. In order to identify potential targets of T cell cross-reactivity, we implemented an in silico strategy combining sequence-based and structure-based methods to screen over 13,5 million possible cross-reactive peptide pairs from BCG and SARS-CoV-2. Our study produced (i) a list of immunogenic BCG-derived peptides that may prime T cell cross-reactivity against SARS-CoV-2, (ii) a large dataset of modeled peptide-HLA structures for the screened targets, and (iii) new computational methods for structure-based screenings that can be used by others in future studies. Our study expands the list of BCG peptides potentially involved in T cell cross-reactivity with SARS-CoV-2-derived peptides, and identifies multiple high-density "neighborhoods" of cross-reactive peptides which could be driving heterologous immunity induced by BCG vaccination, therefore providing insights for future vaccine development efforts.

Markov state modeling reveals alternative unbinding pathways for peptide-MHC complexes.

Peptide binding to major histocompatibility complexes (MHCs) is a central component of the immune system, and understanding the mechanism behind stable peptide-MHC binding will aid the development of immunotherapies. While MHC binding is mostly influenced by the identity of the so-called anchor positions of the peptide, secondary interactions from nonanchor positions are known to play a role in complex stability. However, current MHC-binding prediction methods lack an analysis of the major conformational states and might underestimate the impact of secondary interactions. In this work, we present an atomically detailed analysis of peptide-MHC binding that can reveal the contributions of any interaction toward stability. We propose a simulation framework that uses both umbrella sampling and adaptive sampling to generate a Markov state model (MSM) for a coronavirus-derived peptide (QFKDNVILL), bound to one of the most prevalent MHC receptors in humans (HLA-A24:02). While our model reaffirms the importance of the anchor positions of the peptide in establishing stable interactions, our model also reveals the underestimated importance of position 4 (p4), a nonanchor position. We confirmed our results by simulating the impact of specific peptide mutations and validated these predictions through competitive binding assays. By comparing the MSM of the wild-type system with those of the D4A and D4P mutations, our modeling reveals stark differences in unbinding pathways. The analysis presented here can be applied to any peptide-MHC complex of interest with a structural model as input, representing an important step toward comprehensive modeling of the MHC class I pathway.

Structural Modeling and Molecular Dynamics of the Immune Checkpoint Molecule HLA-G.

HLA-G is considered to be an immune checkpoint molecule, a function that is closely linked to the structure and dynamics of the different HLA-G isoforms. Unfortunately, little is known about the structure and dynamics of these isoforms. For instance, there are only seven crystal structures of HLA-G molecules, being all related to a single isoform, and in some cases lacking important residues associated to the interaction with leukocyte receptors. In addition, they lack information on the dynamics of both membrane-bound HLA-G forms, and soluble forms. We took advantage of in silico strategies to disclose the dynamic behavior of selected HLA-G forms, including the membrane-bound HLA-G1 molecule, soluble HLA-G1 dimer, and HLA-G5 isoform. Both the membrane-bound HLA-G1 molecule and the soluble HLA-G1 dimer were quite stable. Residues involved in the interaction with ILT2 and ILT4 receptors (α3 domain) were very close to the lipid bilayer in the complete HLA-G1 molecule, which might limit accessibility. On the other hand, these residues can be completely exposed in the soluble HLA-G1 dimer, due to the free rotation of the disulfide bridge (Cys42/Cys42). In fact, we speculate that this free rotation of each protomer (i.e., the chains composing the dimer) could enable alternative binding modes for ILT2/ILT4 receptors, which in turn could be associated with greater affinity of the soluble HLA-G1 dimer. Structural analysis of the HLA-G5 isoform demonstrated higher stability for the complex containing the peptide and coupled ß2-microglobulin, while structures lacking such domains were significantly unstable. This study reports for the first time structural conformations for the HLA-G5 isoform and the dynamic behavior of HLA-G1 molecules under simulated biological conditions. All modeled structures were made available through GitHub (https://github.com/KavrakiLab/), enabling their use as templates for modeling other alleles and isoforms, as well as for other computational analyses to investigate key molecular interactions.

Machine Learning-Guided Three-Dimensional Printing of Tissue Engineering Scaffolds.

Various material compositions have been successfully used in 3D printing with promising applications as scaffolds in tissue engineering. However, identifying suitable printing conditions for new materials requires extensive experimentation in a time and resource-demanding process. This study investigates the use of Machine Learning (ML) for distinguishing between printing configurations that are likely to result in low-quality prints and printing configurations that are more promising as a first step toward the development of a recommendation system for identifying suitable printing conditions. The ML-based framework takes as input the printing conditions regarding the material composition and the printing parameters and predicts the quality of the resulting print as either "low" or "high." We investigate two ML-based approaches: a direct classification-based approach that trains a classifier to distinguish between low- and high-quality prints and an indirect approach that uses a regression ML model that approximates the values of a printing quality metric. Both modes are built upon Random Forests. We trained and evaluated the models on a dataset that was generated in a previous study, which investigated fabrication of porous polymer scaffolds by means of extrusion-based 3D printing with a full-factorial design. Our results show that both models were able to correctly label the majority of the tested configurations while a simpler linear ML model was not effective. Additionally, our analysis showed that a full factorial design for data collection can lead to redundancies in the data, in the context of ML, and we propose a more efficient data collection strategy.