Quantification of Uncertainty in Peptide-MHC Binding Prediction Improves High-Affinity Peptide Selection for Therapeutic Design

Neural network extrapolation to distant regions of the protein fitness landscape

Machine learning (ML) has transformed protein engineering by constructing models of the underlying sequence-function landscape to accelerate the discovery of new biomolecules. ML-guided protein design requires models, trained on local sequence-function information, to accurately predict distant fitness peaks. In this work, we evaluate neural networks' capacity to extrapolate beyond their training data. We perform model-guided design using a panel of neural network architectures trained on protein G (GB1)-Immunoglobulin G (IgG) binding data and experimentally test thousands of GB1 designs to systematically evaluate the models' extrapolation. We find each model architecture infers markedly different landscapes from the same data, which give rise to unique design preferences. We find simpler models excel in local extrapolation to design high fitness proteins, while more sophisticated convolutional models can venture deep into sequence space to design proteins that fold but are no longer functional. We also find that implementing a simple ensemble of convolutional neural networks enables robust design of high-performing variants in the local landscape. Our findings highlight how each architecture's inductive biases prime them to learn different aspects of the protein fitness landscape and how a simple ensembling approach makes protein engineering more robust.

Identifying Bayesian optimal experiments for uncertain biochemical pathway models

Pharmacodynamic (PD) models are mathematical models of cellular reaction networks that include drug mechanisms of action. These models are useful for studying predictive therapeutic outcomes of novel drug therapies in silico. However, PD models are known to possess significant uncertainty with respect to constituent parameter data, leading to uncertainty in the model predictions. Furthermore, experimental data to calibrate these models is often limited or unavailable for novel pathways. In this study, we present a Bayesian optimal experimental design approach for improving PD model prediction accuracy. We then apply our method using simulated experimental data to account for uncertainty in hypothetical laboratory measurements. This leads to a probabilistic prediction of drug performance and a quantitative measure of which prospective laboratory experiment will optimally reduce prediction uncertainty in the PD model. The methods proposed here provide a way forward for uncertainty quantification and guided experimental design for models of novel biological pathways.

Transformers meets neoantigen detection: a systematic literature review

Cancer immunology offers a new alternative to traditional cancer treatments, such as radiotherapy and chemotherapy. One notable alternative is the development of personalized vaccines based on cancer neoantigens. Moreover, Transformers are considered a revolutionary development in artificial intelligence with a significant impact on natural language processing (NLP) tasks and have been utilized in proteomics studies in recent years. In this context, we conducted a systematic literature review to investigate how Transformers are applied in each stage of the neoantigen detection process. Additionally, we mapped current pipelines and examined the results of clinical trials involving cancer vaccines.

Artificial intelligence and neoantigens: paving the path for precision cancer immunotherapy

Cancer immunotherapy has witnessed rapid advancement in recent years, with a particular focus on neoantigens as promising targets for personalized treatments. The convergence of immunogenomics, bioinformatics, and artificial intelligence (AI) has propelled the development of innovative neoantigen discovery tools and pipelines. These tools have revolutionized our ability to identify tumor-specific antigens, providing the foundation for precision cancer immunotherapy. AI-driven algorithms can process extensive amounts of data, identify patterns, and make predictions that were once challenging to achieve. However, the integration of AI comes with its own set of challenges, leaving space for further research. With particular focus on the computational approaches, in this article we have explored the current landscape of neoantigen prediction, the fundamental concepts behind, the challenges and their potential solutions providing a comprehensive overview of this rapidly evolving field.

Graph-pMHC: graph neural network approach to MHC class II peptide presentation and antibody immunogenicity

Antigen presentation on MHC class II (pMHCII presentation) plays an essential role in the adaptive immune response to extracellular pathogens and cancerous cells. But it can also reduce the efficacy of large-molecule drugs by triggering an anti-drug response. Significant progress has been made in pMHCII presentation modeling due to the collection of large-scale pMHC mass spectrometry datasets (ligandomes) and advances in machine learning. Here, we develop graph-pMHC, a graph neural network approach to predict pMHCII presentation. We derive adjacency matrices for pMHCII using Alphafold2-multimer and address the peptide-MHC binding groove alignment problem with a simple graph enumeration strategy. We demonstrate that graph-pMHC dramatically outperforms methods with suboptimal inductive biases, such as the multilayer-perceptron-based NetMHCIIpan-4.0 (+20.17% absolute average precision). Finally, we create an antibody drug immunogenicity dataset from clinical trial data and develop a method for measuring anti-antibody immunogenicity risk using pMHCII presentation models. Our model increases receiver operating characteristic curve (ROC)-area under the ROC curve (AUC) by 2.57% compared to just filtering peptides by hits in OASis alone for predicting antibody drug immunogenicity.

Forum on immune digital twins: a meeting report

Medical digital twins are computational models of human biology relevant to a given medical condition, which are tailored to an individual patient, thereby predicting the course of disease and individualized treatments, an important goal of personalized medicine. The immune system, which has a central role in many diseases, is highly heterogeneous between individuals, and thus poses a major challenge for this technology. In February 2023, an international group of experts convened for two days to discuss these challenges related to immune digital twins. The group consisted of clinicians, immunologists, biologists, and mathematical modelers, representative of the interdisciplinary nature of medical digital twin development. A video recording of the entire event is available. This paper presents a synopsis of the discussions, brief descriptions of ongoing digital twin projects at different stages of progress. It also proposes a 5-year action plan for further developing this technology. The main recommendations are to identify and pursue a small number of promising use cases, to develop stimulation-specific assays of immune function in a clinical setting, and to develop a database of existing computational immune models, as well as advanced modeling technology and infrastructure.

Structure-aware deep model for MHC-II peptide binding affinity prediction

The prediction of major histocompatibility complex (MHC)-peptide binding affinity is an important branch in immune bioinformatics, especially helpful in accelerating the design of disease vaccines and immunity therapy. Although deep learning-based solutions have yielded promising results on MHC-II molecules in recent years, these methods ignored structure knowledge from each peptide when employing the deep neural network models. Each peptide sequence has its specific combination order, so it is worth considering adding the structural information of the peptide sequence to the deep model training. In this work, we use positional encoding to represent the structural information of peptide sequences and validly combine the positional encoding with existing models by different strategies. Experiments on three datasets show that the introduction of position-coding information can further improve the performance built upon the existing model. The idea of introducing positional encoding to this field can provide important reference significance for the optimization of the deep network structure in the future.

RPEMHC: improved prediction of MHC-peptide binding affinity by a deep learning approach based on residue-residue pair encoding

MOTIVATION

Binding of peptides to major histocompatibility complex (MHC) molecules plays a crucial role in triggering T cell recognition mechanisms essential for immune response. Accurate prediction of MHC-peptide binding is vital for the development of cancer therapeutic vaccines. While recent deep learning-based methods have achieved significant performance in predicting MHC-peptide binding affinity, most of them separately encode MHC molecules and peptides as inputs, potentially overlooking critical interaction information between the two.

RESULTS

In this work, we propose RPEMHC, a new deep learning approach based on residue-residue pair encoding to predict the binding affinity between peptides and MHC, which encode an MHC molecule and a peptide as a residue-residue pair map. We evaluate the performance of RPEMHC on various MHC-II-related datasets for MHC-peptide binding prediction, demonstrating that RPEMHC achieves better or comparable performance against other state-of-the-art baselines. Moreover, we further construct experiments on MHC-I-related datasets, and experimental results demonstrate that our method can work on both two MHC classes. These extensive validations have manifested that RPEMHC is an effective tool for studying MHC-peptide interactions and can potentially facilitate the vaccine development.

AVAILABILITY

The source code of the method along with trained models is freely available at https://github.com/lennylv/RPEMHC.

Calibrated geometric deep learning improves kinase-drug binding predictions

Protein kinases regulate various cellular functions and hold significant pharmacological promise in cancer and other diseases. Although kinase inhibitors are one of the largest groups of approved drugs, much of the human kinome remains unexplored but potentially druggable. Computational approaches, such as machine learning, offer efficient solutions for exploring kinase-compound interactions and uncovering novel binding activities. Despite the increasing availability of three-dimensional (3D) protein and compound structures, existing methods predominantly focus on exploiting local features from one-dimensional protein sequences and two-dimensional molecular graphs to predict binding affinities, overlooking the 3D nature of the binding process. Here we present KDBNet, a deep learning algorithm that incorporates 3D protein and molecule structure data to predict binding affinities. KDBNet uses graph neural networks to learn structure representations of protein binding pockets and drug molecules, capturing the geometric and spatial characteristics of binding activity. In addition, we introduce an algorithm to quantify and calibrate the uncertainties of KDBNet's predictions, enhancing its utility in model-guided discovery in chemical or protein space. Experiments demonstrated that KDBNet outperforms existing deep learning models in predicting kinase-drug binding affinities. The uncertainties estimated by KDBNet are informative and well-calibrated with respect to prediction errors. When integrated with a Bayesian optimization framework, KDBNet enables data-efficient active learning and accelerates the exploration and exploitation of diverse high-binding kinase-drug pairs.

Neural network extrapolation to distant regions of the protein fitness landscape

Machine learning (ML) has transformed protein engineering by constructing models of the underlying sequence-function landscape to accelerate the discovery of new biomolecules. ML-guided protein design requires models, trained on local sequence-function information, to accurately predict distant fitness peaks. In this work, we evaluate neural networks' capacity to extrapolate beyond their training data. We perform model-guided design using a panel of neural network architectures trained on protein G (GB1)-Immunoglobulin G (IgG) binding data and experimentally test thousands of GB1 designs to systematically evaluate the models' extrapolation. We find each model architecture infers markedly different landscapes from the same data, which give rise to unique design preferences. We find simpler models excel in local extrapolation to design high fitness proteins, while more sophisticated convolutional models can venture deep into sequence space to design proteins that fold but are no longer functional. Our findings highlight how each architecture's inductive biases prime them to learn different aspects of the protein fitness landscape.

Accelerated NLRP3 inflammasome-inhibitory peptide design using a recurrent neural network model and molecular dynamics simulations

Anomalous NLRP3 inflammasome responses have been linked to multiple health issues, including but not limited to atherosclerosis, diabetes, metabolic syndrome, cardiovascular disease, and neurodegenerative disease. Thus, targeting NLRP3 and modulating its associated immune response might be a promising strategy for developing new anti-inflammatory drugs. Herein, we report a computational method for de novo peptide design for targeting NLRP3 inflammasomes. The described method leverages a long-short-term memory (LSTM) network based on a recurrent neural network (RNN) to model a valuable latent space of molecules. The resulting classifiers are utilized to guide the selection of molecules generated by the model based on circular dichroism spectra and physicochemical features derived from high-throughput molecular dynamics simulations. Of the experimentally tested sequences, 60% of the peptides showed NLRP3-mediated inhibition of IL-1β and IL-18. One peptide displayed high potency against NLRP3-mediated IL-1β inhibition. However, NLRC4 and AIM2 inflammasome-mediated IL-1β secretion was uninterrupted by this peptide, demonstrating its selectivity toward the NLRP3 inflammasome. Overall, these results indicate that deep learning and molecular dynamics can accelerate the discovery of NLRP3 inhibitors with potent and selective activity.

TransMHCII: a novel MHC-II binding prediction model built using a protein language model and an image classifier

The emergence of deep learning models such as AlphaFold2 has revolutionized the structure prediction of proteins. Nevertheless, much remains unexplored, especially on how we utilize structure models to predict biological properties. Herein, we present a method using features extracted from protein language models (PLMs) to predict the major histocompatibility complex class II (MHC-II) binding affinity of peptides. Specifically, we evaluated a novel transfer learning approach where the backbone of our model was interchanged with architectures designed for image classification tasks. Features extracted from several PLMs (ESM1b, ProtXLNet or ProtT5-XL-UniRef) were passed into image models (EfficientNet v2b0, EfficientNet v2m or ViT-16). The optimal pairing of the PLM and image classifier resulted in the final model TransMHCII, outperforming NetMHCIIpan 3.2 and NetMHCIIpan 4.0-BA on the receiver operating characteristic area under the curve, balanced accuracy and Jaccard scores. The architecture innovation may facilitate the development of other deep learning models for biological problems.

Align-gram: Rethinking the Skip-gram Model for Protein Sequence Analysis

The inception of next generations sequencing technologies have exponentially increased the volume of biological sequence data. Protein sequences, being quoted as the 'language of life', has been analyzed for a multitude of applications and inferences. Owing to the rapid development of deep learning, in recent years there have been a number of breakthroughs in the domain of Natural Language Processing. Since these methods are capable of performing different tasks when trained with a sufficient amount of data, off-the-shelf models are used to perform various biological applications. In this study, we investigated the applicability of the popular Skip-gram model for protein sequence analysis and made an attempt to incorporate some biological insights into it. We propose a novel k-mer embedding scheme, Align-gram, which is capable of mapping the similar k-mers close to each other in a vector space. Furthermore, we experiment with other sequence-based protein representations and observe that the embeddings derived from Align-gram aids modeling and training deep learning models better. Our experiments with a simple baseline LSTM model and a much complex CNN model of DeepGoPlus shows the potential of Align-gram in performing different types of deep learning applications for protein sequence analysis.

Application of machine learning on understanding biomolecule interactions in cellular machinery

Machine learning (ML) applications have become ubiquitous in all fields of research including protein science and engineering. Apart from protein structure and mutation prediction, scientists are focusing on knowledge gaps with respect to the molecular mechanisms involved in protein binding and interactions with other components in the experimental setups or the human body. Researchers are working on several wet-lab techniques and generating data for a better understanding of concepts and mechanics involved. The information like biomolecular structure, binding affinities, structure fluctuations and movements are enormous which can be handled and analyzed by ML. Therefore, this review highlights the significance of ML in understanding the biomolecular interactions while assisting in various fields of research such as drug discovery, nanomedicine, nanotoxicity and material science. Hence, the way ahead would be to force hand-in hand of laboratory work and computational techniques.

Attentive Variational Information Bottleneck for TCR-peptide interaction prediction

MOTIVATION

We present a multi-sequence generalization of Variational Information Bottleneck and call the resulting model Attentive Variational Information Bottleneck (AVIB). Our AVIB model leverages multi-head self-attention to implicitly approximate a posterior distribution over latent encodings conditioned on multiple input sequences. We apply AVIB to a fundamental immuno-oncology problem: predicting the interactions between T-cell receptors (TCRs) and peptides.

RESULTS

Experimental results on various datasets show that AVIB significantly outperforms state-of-the-art methods for TCR-peptide interaction prediction. Additionally, we show that the latent posterior distribution learned by AVIB is particularly effective for the unsupervised detection of out-of-distribution amino acid sequences.

AVAILABILITY AND IMPLEMENTATION

The code and the data used for this study are publicly available at: https://github.com/nec-research/vibtcr.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Conformal prediction under feedback covariate shift for biomolecular design

Many applications of machine-learning methods involve an iterative protocol in which data are collected, a model is trained, and then outputs of that model are used to choose what data to consider next. For example, a data-driven approach for designing proteins is to train a regression model to predict the fitness of protein sequences and then use it to propose new sequences believed to exhibit greater fitness than observed in the training data. Since validating designed sequences in the wet laboratory is typically costly, it is important to quantify the uncertainty in the model's predictions. This is challenging because of a characteristic type of distribution shift between the training and test data that arises in the design setting-one in which the training and test data are statistically dependent, as the latter is chosen based on the former. Consequently, the model's error on the test data-that is, the designed sequences-has an unknown and possibly complex relationship with its error on the training data. We introduce a method to construct confidence sets for predictions in such settings, which account for the dependence between the training and test data. The confidence sets we construct have finite-sample guarantees that hold for any regression model, even when it is used to choose the test-time input distribution. As a motivating use case, we use real datasets to demonstrate how our method quantifies uncertainty for the predicted fitness of designed proteins and can therefore be used to select design algorithms that achieve acceptable tradeoffs between high predicted fitness and low predictive uncertainty.

On TCR binding predictors failing to generalize to unseen peptides

Several recent studies investigate TCR-peptide/-pMHC binding prediction using machine learning or deep learning approaches. Many of these methods achieve impressive results on test sets, which include peptide sequences that are also included in the training set. In this work, we investigate how state-of-the-art deep learning models for TCR-peptide/-pMHC binding prediction generalize to unseen peptides. We create a dataset including positive samples from IEDB, VDJdb, McPAS-TCR, and the MIRA set, as well as negative samples from both randomization and 10X Genomics assays. We name this collection of samples TChard. We propose the hard split, a simple heuristic for training/test split, which ensures that test samples exclusively present peptides that do not belong to the training set. We investigate the effect of different training/test splitting techniques on the models’ test performance, as well as the effect of training and testing the models using mismatched negative samples generated randomly, in addition to the negative samples derived from assays. Our results show that modern deep learning methods fail to generalize to unseen peptides. We provide an explanation why this happens and verify our hypothesis on the TChard dataset. We then conclude that robust prediction of TCR recognition is still far for being solved.

A high-throughput yeast display approach to profile pathogen proteomes for MHC-II binding

T cells play a critical role in the adaptive immune response, recognizing peptide antigens presented on the cell surface by major histocompatibility complex (MHC) proteins. While assessing peptides for MHC binding is an important component of probing these interactions, traditional assays for testing peptides of interest for MHC binding are limited in throughput. Here, we present a yeast display-based platform for assessing the binding of tens of thousands of user-defined peptides in a high-throughput manner. We apply this approach to assess a tiled library covering the SARS-CoV-2 proteome and four dengue virus serotypes for binding to human class II MHCs, including HLA-DR401, -DR402, and -DR404. While the peptide datasets show broad agreement with previously described MHC-binding motifs, they additionally reveal experimentally validated computational false positives and false negatives. We therefore present this approach as able to complement current experimental datasets and computational predictions. Further, our yeast display approach underlines design considerations for epitope identification experiments and serves as a framework for examining relationships between viral conservation and MHC binding, which can be used to identify potentially high-interest peptide binders from viral proteins. These results demonstrate the utility of our approach to determine peptide-MHC binding interactions in a manner that can supplement and potentially enhance current algorithm-based approaches.

The role of neoantigens in tumor immunotherapy

Tumor neoantigens are aberrant polypeptides produced by tumor cells as a result of genomic mutations. They are also tumor-specific antigens (TSA). Neoantigens are more immunogenic than tumor-related antigens and do not induce autoimmunity. Based on the rapid development of bioinformatics and the continuous update of sequencing technology, cancer immunotherapy with tumor neoantigens has made promising breakthroughs and progress. In this review, the generation, prediction, and identification of novel antigens, as well as the individualized treatments of neoantigens, were first introduced. Secondly, the mechanism of Chimeric Antigen Receptor T-Cell Immunotherapy (CAR-T) therapy and immune checkpoint blockade therapy in the treatment of tumors were outlined, and the three treatment methods were compared. Thirdly, the application of neoantigens in CAR-T therapy and PD-1/PD-L1 blockade therapy was briefly described. The benefits of the neoantigen vaccines over common vaccines were summarized as well. Finally, the prospect of neoantigen therapy was presented.

OUP accepted manuscript

Motivation

Computationally predicting major histocompatibility complex (MHC)-peptide binding affinity is an important problem in immunological bioinformatics. Recent cutting-edge deep learning-based methods for this problem are unable to achieve satisfactory performance for MHC class II molecules. This is because such methods generate the input by simply concatenating the two given sequences: (the estimated binding core of) a peptide and (the pseudo sequence of) an MHC class II molecule, ignoring biological knowledge behind the interactions of the two molecules. We thus propose a binding core-aware deep learning-based model, DeepMHCII, with a binding interaction convolution layer, which allows to integrate all potential binding cores (in a given peptide) with the MHC pseudo (binding) sequence, through modeling the interaction with multiple convolutional kernels.

Results

Extensive empirical experiments with four large-scale datasets demonstrate that DeepMHCII significantly outperformed four state-of-the-art methods under numerous settings, such as 5-fold cross-validation, leave one molecule out, validation with independent testing sets and binding core prediction. All these results and visualization of the predicted binding cores indicate the effectiveness of our model, DeepMHCII, and the importance of properly modeling biological facts in deep learning for high predictive performance and efficient knowledge discovery.

Availability and implementation

DeepMHCII is publicly available at https://github.com/yourh/DeepMHCII.

Supplementary information

Supplementary data are available at Bioinformatics online.

Adaptive machine learning for protein engineering

Machine-learning models that learn from data to predict how protein sequence encodes function are emerging as a useful protein engineering tool. However, when using these models to suggest new protein designs, one must deal with the vast combinatorial complexity of protein sequences. Here, we review how to use a sequence-to-function machine-learning surrogate model to select sequences for experimental measurement. First, we discuss how to select sequences through a single round of machine-learning optimization. Then, we discuss sequential optimization, where the goal is to discover optimized sequences and improve the model across multiple rounds of training, optimization, and experimental measurement.

Technological advances in cancer immunity: from immunogenomics to single-cell analysis and artificial intelligence

Immunotherapies play critical roles in cancer treatment. However, given that only a few patients respond to immune checkpoint blockades and other immunotherapeutic strategies, more novel technologies are needed to decipher the complicated interplay between tumor cells and the components of the tumor immune microenvironment (TIME). Tumor immunomics refers to the integrated study of the TIME using immunogenomics, immunoproteomics, immune-bioinformatics, and other multi-omics data reflecting the immune states of tumors, which has relied on the rapid development of next-generation sequencing. High-throughput genomic and transcriptomic data may be utilized for calculating the abundance of immune cells and predicting tumor antigens, referring to immunogenomics. However, as bulk sequencing represents the average characteristics of a heterogeneous cell population, it fails to distinguish distinct cell subtypes. Single-cell-based technologies enable better dissection of the TIME through precise immune cell subpopulation and spatial architecture investigations. In addition, radiomics and digital pathology-based deep learning models largely contribute to research on cancer immunity. These artificial intelligence technologies have performed well in predicting response to immunotherapy, with profound significance in cancer therapy. In this review, we briefly summarize conventional and state-of-the-art technologies in the field of immunogenomics, single-cell and artificial intelligence, and present prospects for future research.

MHCVision: estimation of global and local false discovery rate for MHC class I peptide binding prediction

Motivation

MHC-peptide binding prediction has been widely used for understanding the immune response of individuals or populations, each carrying different MHC molecules as well as for the development of immunotherapeutics. The results from MHC-peptide binding prediction tools are mostly reported as a predicted binding affinity (IC₅₀) and the percentile rank score, and global thresholds e.g. IC₅₀ value < 500 nM or percentile rank < 2% are generally recommended for distinguishing binding peptides from non-binding peptides. However, it is difficult to evaluate statistically the probability of an individual peptide binding prediction to be true or false solely considering predicted scores. Therefore, statistics describing the overall global false discovery rate (FDR) and local FDR, also called posterior error probability (PEP) are required to give statistical context to the natively produced scores.

Result

We have developed an algorithm and code implementation, called MHCVision, for estimation of FDR and PEP values for the predicted results of MHC-peptide binding prediction from the NetMHCpan tool. MHCVision performs parameter estimation using a modified expectation maximization framework for a two-component beta mixture model, representing the distribution of true and false scores of the predicted dataset. We can then estimate the PEP of an individual peptide’s predicted score, and conversely the probability that it is true. We demonstrate that the use of global FDR and PEP estimation can provide a better trade-off between sensitivity and precision over using currently recommended thresholds from tools.

Availability and implementation

https://github.com/PGB-LIV/MHCVision.

Supplementary information

Supplementary data are available at Bioinformatics online.

Learning the protein language: Evolution, structure, and function

Language models have recently emerged as a powerful machine-learning approach for distilling information from massive protein sequence databases. From readily available sequence data alone, these models discover evolutionary, structural, and functional organization across protein space. Using language models, we can encode amino-acid sequences into distributed vector representations that capture their structural and functional properties, as well as evaluate the evolutionary fitness of sequence variants. We discuss recent advances in protein language modeling and their applications to downstream protein property prediction problems. We then consider how these models can be enriched with prior biological knowledge and introduce an approach for encoding protein structural knowledge into the learned representations. The knowledge distilled by these models allows us to improve downstream function prediction through transfer learning. Deep protein language models are revolutionizing protein biology. They suggest new ways to approach protein and therapeutic design. However, further developments are needed to encode strong biological priors into protein language models and to increase their accessibility to the broader community.

BERTMHC: Improved MHC-peptide class II interaction prediction with transformer and multiple instance learning

Motivation

Increasingly comprehensive characterization of cancer-associated genetic alterations has paved the way for the development of highly specific therapeutic vaccines. Predicting precisely the binding and presentation of peptides to major histocompatibility complex (MHC) alleles is an important step toward such therapies. Recent data suggest that presentation of both class I and II epitopes are critical for the induction of a sustained effective immune response. However, the prediction performance for MHC class II has been limited compared to class I.

Results

We present a transformer neural network model which leverages self-supervised pretraining from a large corpus of protein sequences. We also propose a multiple instance learning (MIL) framework to deconvolve mass spectrometry data where multiple potential MHC alleles may have presented each peptide. We show that pretraining boosted the performance for these tasks. Combining pretraining and the novel MIL approach, our model outperforms state-of-the-art models based on peptide and MHC sequence only for both binding and cell surface presentation predictions.

Availability and implementation

Our source code is available at https://github.com/s6juncheng/BERTMHC under a noncommercial license. A webserver is available at https://bertmhc.privacy.nlehd.de/

Supplementary information

Supplementary data are available at Bioinformatics online.

Machine learning optimization of peptides for presentation by class II MHCs

T cells play a critical role in cellular immune responses to pathogens and cancer and can be activated and expanded by MHC-presented antigens contained in peptide vaccines. We present a machine learning method to optimize the presentation of peptides by class II MHCs by modifying their anchor residues. Our method first learns a model of peptide affinity for a class II MHC using an ensemble of deep residual networks, and then uses the model to propose anchor residue changes to improve peptide affinity. We use a high throughput yeast display assay to show that anchor residue optimization improves peptide binding. Supplementary information: Supplementary data are available at Bioinformatics online.

MHCAttnNet: predicting MHC-peptide bindings for MHC alleles classes I and II using an attention-based deep neural model

Motivation

Accurate prediction of binding between a major histocompatibility complex (MHC) allele and a peptide plays a major role in the synthesis of personalized cancer vaccines. The immune system struggles to distinguish between a cancerous and a healthy cell. In a patient suffering from cancer who has a particular MHC allele, only those peptides that bind with the MHC allele with high affinity, help the immune system recognize the cancerous cells.

Results

MHCAttnNet is a deep neural model that uses an attention mechanism to capture the relevant subsequences of the amino acid sequences of peptides and MHC alleles. It then uses this to accurately predict the MHC-peptide binding. MHCAttnNet achieves an AUC-PRC score of 94.18% with 161 class I MHC alleles, which outperforms the state-of-the-art models for this task. MHCAttnNet also achieves a better F1-score in comparison to the state-of-the-art models while covering a larger number of class II MHC alleles. The attention mechanism used by MHCAttnNet provides a heatmap over the amino acids thus indicating the important subsequences present in the amino acid sequence. This approach also allows us to focus on a much smaller number of relevant trigrams corresponding to the amino acid sequence of an MHC allele, from 9251 possible trigrams to about 258. This significantly reduces the number of amino acid subsequences that need to be clinically tested.

Availability and implementation

The data and source code are available at https://github.com/gopuvenkat/MHCAttnNet.

COVID-19: The Effect of Host Genetic Variations on Host-Virus Interactions

Spurred into action by the COVID-19 pandemic, the global scientific community has, in a short of period of time, made astonishing progress in understanding and combating COVID-19. Given the known human protein machinery for (a) SARS-CoV-2 entry, (b) the host innate immune response, and (c) virus-host interactions (protein-protein and RNA-protein), the potential effects of human genetic variation in this machinery, which may contribute to clinical differences in SARS-CoV-2 pathogenesis and help determine individual risk for COVID-19 infection, are explored. The Genome Aggregation Database (gnomAD) was used to show that several rare germline exome variants of proteins in these pathways occur in the human population, suggesting that carriers of these rare variants (especially for proteins of innate immunity pathways) are at risk for severe symptoms (like the severe symptoms in patients who are known to be rare variant carriers), whereas carriers of other variants could have a protective advantage against infection. The occurrence of genetic variation is thus expected to motivate the experimental probing of natural variants to understand the mechanistic differences in SARS-CoV-2 pathogenesis from one individual to another.

Predicted Cellular Immunity Population Coverage Gaps for SARS-CoV-2 Subunit Vaccines and Their Augmentation by Compact Peptide Sets

Subunit vaccines induce immunity to a pathogen by presenting a component of the pathogen and thus inherently limit the representation of pathogen peptides for cellular immunity-based memory. We find that severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) subunit peptides may not be robustly displayed by the major histocompatibility complex (MHC) molecules in certain individuals. We introduce an augmentation strategy for subunit vaccines that adds a small number of SARS-CoV-2 peptides to a vaccine to improve the population coverage of pathogen peptide display. Our population coverage estimates integrate clinical data on peptide immunogenicity in convalescent COVID-19 patients and machine learning predictions. We evaluate the population coverage of 9 different subunits of SARS-CoV-2, including 5 functional domains and 4 full proteins, and augment each of them to fill a predicted coverage gap.

•

Clinical data and machine learning predict SARS-CoV-2 peptide-HLA immunogenicity

•

Human population coverage gaps of COVID-19 subunit vaccines are predicted

•

Subunit augmentation improves vaccine population coverage for cellular immunity

•

Subunit-free peptide vaccines are predicted to have high population coverage

Deep Learning in Proteomics

Proteomics, the study of all the proteins in biological systems, is becoming a data-rich science. Protein sequences and structures are comprehensively catalogued in online databases. With recent advancements in tandem mass spectrometry (MS) technology, protein expression and post-translational modifications (PTMs) can be studied in a variety of biological systems at the global scale. Sophisticated computational algorithms are needed to translate the vast amount of data into novel biological insights. Deep learning automatically extracts data representations at high levels of abstraction from data, and it thrives in data-rich scientific research domains. Here, a comprehensive overview of deep learning applications in proteomics, including retention time prediction, MS/MS spectrum prediction, de novo peptide sequencing, PTM prediction, major histocompatibility complex-peptide binding prediction, and protein structure prediction, is provided. Limitations and the future directions of deep learning in proteomics are also discussed. This review will provide readers an overview of deep learning and how it can be used to analyze proteomics data.

Leveraging Uncertainty in Machine Learning Accelerates Biological Discovery and Design

Machine learning that generates biological hypotheses has transformative potential, but most learning algorithms are susceptible to pathological failure when exploring regimes beyond the training data distribution. A solution to address this issue is to quantify prediction uncertainty so that algorithms can gracefully handle novel phenomena that confound standard methods. Here, we demonstrate the broad utility of robust uncertainty prediction in biological discovery. By leveraging Gaussian process-based uncertainty prediction on modern pre-trained features, we train a model on just 72 compounds to make predictions over a 10,833-compound library, identifying and experimentally validating compounds with nanomolar affinity for diverse kinases and whole-cell growth inhibition of Mycobacterium tuberculosis. Uncertainty facilitates a tight iterative loop between computation and experimentation and generalizes across biological domains as diverse as protein engineering and single-cell transcriptomics. More broadly, our work demonstrates that uncertainty should play a key role in the increasing adoption of machine learning algorithms into the experimental lifecycle.

Computationally Optimized SARS-CoV-2 MHC Class I and II Vaccine Formulations Predicted to Target Human Haplotype Distributions

We present a combinatorial machine learning method to evaluate and optimize peptide vaccine formulations for SARS-CoV-2. Our approach optimizes the presentation likelihood of a diverse set of vaccine peptides conditioned on a target human-population HLA haplotype distribution and expected epitope drift. Our proposed SARS-CoV-2 MHC class I vaccine formulations provide 93.21% predicted population coverage with at least five vaccine peptide-HLA average hits per person (≥ 1 peptide: 99.91%) with all vaccine peptides perfectly conserved across 4,690 geographically sampled SARS-CoV-2 genomes. Our proposed MHC class II vaccine formulations provide 97.21% predicted coverage with at least five vaccine peptide-HLA average hits per person with all peptides having an observed mutation probability of ≤ 0.001. We provide an open-source implementation of our design methods (OptiVax), vaccine evaluation tool (EvalVax), as well as the data used in our design efforts here: https://github.com/gifford-lab/optivax.

Uncovering the Tumor Antigen Landscape: What to Know about the Discovery Process

According to the latest available data, cancer is the second leading cause of death, highlighting the need for novel cancer therapeutic approaches. In this context, immunotherapy is emerging as a reliable first-line treatment for many cancers, particularly metastatic melanoma. Indeed, cancer immunotherapy has attracted great interest following the recent clinical approval of antibodies targeting immune checkpoint molecules, such as PD-1, PD-L1, and CTLA-4, that release the brakes of the immune system, thus reviving a field otherwise poorly explored. Cancer immunotherapy mainly relies on the generation and stimulation of cytotoxic CD8 T lymphocytes (CTLs) within the tumor microenvironment (TME), priming T cells and establishing efficient and durable anti-tumor immunity. Therefore, there is a clear need to define and identify immunogenic T cell epitopes to use in therapeutic cancer vaccines. Naturally presented antigens in the human leucocyte antigen-1 (HLA-I) complex on the tumor surface are the main protagonists in evocating a specific anti-tumor CD8+ T cell response. However, the methodologies for their identification have been a major bottleneck for their reliable characterization. Consequently, the field of antigen discovery has yet to improve. The current review is intended to define what are today known as tumor antigens, with a main focus on CTL antigenic peptides. We also review the techniques developed and employed to date for antigen discovery, exploring both the direct elution of HLA-I peptides and the in silico prediction of epitopes. Finally, the last part of the review analyses the future challenges and direction of the antigen discovery field.

Robust computational design and evaluation of peptide vaccines for cellular immunity with application to SARS-CoV-2

We present a combinatorial machine learning method to evaluate and optimize peptide vaccine formulations, and we find for SARS-CoV-2 that it provides superior predicted display of viral epitopes by MHC class I and MHC class II molecules over populations when compared to other candidate vaccines. Our method is robust to idiosyncratic errors in the prediction of MHC peptide display and considers target population HLA haplotype frequencies during optimization. To minimize clinical development time our methods validate vaccines with multiple peptide presentation algorithms to increase the probability that a vaccine will be effective. We optimize an objective function that is based on the presentation likelihood of a diverse set of vaccine peptides conditioned on a target population HLA haplotype distribution and expected epitope drift. We produce separate peptide formulations for MHC class I loci (HLA-A, HLA-B, and HLA-C) and class II loci (HLA-DP, HLA-DQ, and HLA-DR) to permit signal sequence based cell compartment targeting using nucleic acid based vaccine platforms. Our SARS-CoV-2 MHC class I vaccine formulations provide 93.21% predicted population coverage with at least five vaccine peptide-HLA hits on average in an individual (≥ 1 peptide 99.91%) with all vaccine peptides perfectly conserved across 4,690 geographically sampled SARS-CoV-2 genomes. Our MHC class II vaccine formulations provide 90.17% predicted coverage with at least five vaccine peptide-HLA hits on average in an individual with all peptides having observed mutation probability ≤ 0.001. We evaluate 29 previously published peptide vaccine designs with our evaluation tool with the requirement of having at least five vaccine peptide-HLA hits per individual, and they have a predicted maximum of 58.51% MHC class I coverage and 71.65% MHC class II coverage given haplotype based analysis. We provide an open source implementation of our design methods (OptiVax), vaccine evaluation tool (EvalVax), as well as the data used in our design efforts.

Peptide-Major Histocompatibility Complex Class I Binding Prediction Based on Deep Learning With Novel Feature

Peptide-based vaccine development needs accurate prediction of the binding affinity between major histocompatibility complex I (MHC I) proteins and their peptide ligands. Nowadays more and more machine learning methods have been developed to predict binding affinity and some of them have become the popular tools. However most of them are designed by the shallow neural networks. Bengio said that deep neural networks can learn better fits with less data than shallow neural networks. In our case, some of the alleles only have dozens of peptide data. In addition, we transform each peptide into a characteristic matrix and input it into the model. As we know when dealing with the problem that the input is a matrix, convolutional neural network (CNN) can find the most critical features by itself. Obviously, compared with the traditional neural network model, CNN is more suitable for predicting binding affinity. Different from the previous studies which are based on blocks substitution matrix (BLOSUM), we used novel feature to do the prediction. Since we consider that the order of the sequence, hydropathy index, polarity and the length of the peptide could affect the binding affinity and the properties of these amino acids are key factors for their binding to MHC, we extracted these information from each peptide. In order to make full use of the data we have obtained, we have integrated different lengths of peptides into 15mer based on the binding mode of peptide to MHC I. In order to demonstrate that our method is reliable to predict peptide-MHC binding, we compared our method with several popular methods. The experiments show the superiority of our method.