DeepMHCII: A Novel Binding Core-Aware Deep Interaction Model for Accurate MHC II-peptide Binding Affinity Prediction

Computationally predicting 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 the biological knowledge behind the interactions of the two molecules. We thus propose a binding core-aware deep learning-based model, DeepMHCII, with binding interaction convolution layer (BICL), which allows integrating all potential binding cores (in a given peptide) and the MHC pseudo (binding) sequence, through modeling the interaction with multiple convolutional kernels. Extensive empirical experiments with four large-scale datasets demonstrate that DeepMHCII significantly outperformed four state-of-the-art methods under numerous settings, such as five-fold cross-validation, leave one molecule out, validation with independent testing sets, and binding core prediction. All these results with visualization of the predicted binding cores indicate the effectiveness and importance of properly modeling biological facts in deep learning for high performance and knowledge discovery. DeepMHCII is publicly available at https://weilab.sjtu.edu.cn/DeepMHCII/.

485 RAFT: A framework to support rapid and reproducible immuno-oncology analyses

BackgroundAnalysis reproducibility and transparency are pillars of robust and trustworthy scientific results. The dependability of these results is crucial in clinical settings where they may guide high-impact decisions affecting patient health. Independent reproduction of computational results has been problematic and can be a burden on the individuals attempting to reproduce the results. Reproduction complications may arise from: 1) insufficiently described parameters, 2) vague methods, or 3) secret scripts required to generate final outputs, among others. Here we introduce RAFT (Reproducible Analyses Framework and Tools), a framework for immuno-oncology biomarker development built with Python 3 and Nextflow DSL2 which aims to enable end-to-end reproducibility of entire computational analyses in multiple contexts (e.g. local, compute cluster, or cloud) with minimal overhead through a focus on usability (figures 1 and 2).MethodsRAFT builds upon Nextflow’s DSL2 module-based approach to workflows by providing a ‘project’ context upon which users can add metadata, load references, and build up their analysis step-by-step. RAFT also has pre-built modules with workflows commonly utilized in immuno-oncology analyses (e.g. TCR/BCR repertoire reconstruction and HLA typing) and aids users through automatic module dependency resolution. Transparency is gained by having a single end-to-end script containing all steps and parameters as well as a single configuration file. Finally, RAFT allows users to create and share a package of project metadata files including the main script, all input and output checksums, all modules, and the RAFT steps required to create the analysis. This package, coupled with any required inputs files, can be used to recreate the analysis or further expand an analysis with additional datasets or alternative parameters.ResultsRAFT has been used by our computational team to create an immuno-oncology meta-analysis submitted to SITC 2020. A simple, proof-of-concept analysis has been used to establish RAFT’s ability to support reproducibility by running locally on laptop computers, on multiple research compute clusters, and on the Google Cloud Platform.Abstract 485 Figure 1Example RAFT UsageUsers define their required inputs, build their analysis, and run their analysis using the RAFT command-line interface. The metadata from the analysis can then be shared through a RAFT package with collaborators or interested third-parties in order to reproduce or expand upon the initial results.Abstract 485 Figure 2End-to-end RAFTRAFT supports end-to-end analysis development through a ‘project’ structure. Users link local required files (e.g. FASTQs, references or manifests) into their appropriate/raft subdirectory. (1) Projects are initiated using the raft init-project command which creates and populates a project-specific directory. (2–3) Users then load required metadata (e.g. sample manifests or clinical data) and references (e.g. alignment references) into the project using the raft load-metadata or raft load-reference commands, respectively. (4) Modules consisting of tool-specific and topical workflows are cloned from a collection of remote repositories into the project using raft load-module. (5) Specific processes and workflows from previously loaded modules are added to the analysis (main.nf) through raft add-step. Users can then modify main.nf with their desired parameters and execute the workflow using raft run-workflow. (6) Additionally, RAFT allows an iterative approach where results from RAFT can be analyzed and modified through RStudio and re-run through Nextflow.ConclusionsThe RAFT platform shows promising capabilities to support rapid and reproducible research within the field of immuno-oncology. Several features remain in development and testing, such as incorporation of additional immunogenomics feature modules such as variant/fusion detection and HLA/peptide binding affinity estimation. Other functionality in development will enable collaborators to use remote Git repository hosting (e.g. GitHub or GitLab) to jointly and iteratively modify an analysis.

Mutation position is an important determinant for predicting cancer neoantigens

Tumor-specific mutations can generate neoantigens that drive CD8 T cell responses against cancer. Next-generation sequencing and computational methods have been successfully applied to identify mutations and predict neoantigens. However, only a small fraction of predicted neoantigens are immunogenic. Currently, predicted peptide binding affinity for MHC-I is often the major criterion for prioritizing neoantigens, although little progress has been made toward understanding the precise functional relationship between affinity and immunogenicity. We therefore systematically assessed the immunogenicity of peptides containing single amino acid mutations in mouse tumor models and divided them into two classes of immunogenic mutations. The first comprises mutations at a nonanchor residue, for which we find that the predicted absolute binding affinity is predictive of immunogenicity. The second involves mutations at an anchor residue; here, predicted relative affinity (compared with the WT counterpart) is a better predictor. Incorporating these features into an immunogenicity model significantly improves neoantigen ranking. Importantly, these properties of neoantigens are also predictive in human datasets, suggesting that they can be used to prioritize neoantigens for individualized neoantigen-specific immunotherapies.

Machine Learning in Quantitative Protein–peptide Affinity Prediction: Implications for Therapeutic Peptide Design

Background:Protein–peptide recognition plays an essential role in the orchestration and regulation of cell signaling networks, which is estimated to be responsible for up to 40% of biological interaction events in the human interactome and has recently been recognized as a new and attractive druggable target for drug development and disease intervention.Methods:We present a systematic review on the application of machine learning techniques in the quantitative modeling and prediction of protein–peptide binding affinity, particularly focusing on its implications for therapeutic peptide design. We also briefly introduce the physical quantities used to characterize protein–peptide affinity and attempt to extend the content of generalized machine learning methods.Results:Existing issues and future perspective on the statistical modeling and regression prediction of protein– peptide binding affinity are discussed.Conclusion:There is still a long way to go before establishment of general, reliable and efficient machine leaningbased protein–peptide affinity predictors.

Peptide binding affinity redistributes preassembled repeat protein fragments

Designed armadillo repeat proteins (dArmRPs) are modular peptide binders composed of N- and C-terminal capping repeats Y and A and a variable number of internal modules M that each specifically recognize two amino acids of the target peptide. Complementary fragments of dArmRPs obtained by splitting the protein between helices H1 and H2 of an internal module show conditional and specific assembly only in the presence of a target peptide (Michel, E., Plückthun, A., and Zerbe, O. (2018). Peptide-guided assembly of repeat protein fragments. Angew. Chem. Int. Ed. 57, 4576–4579). Here, we investigate dArmRP fragments that already spontaneously assemble with high affinity, e.g. those obtained from splits between entire modules or between helices H2 and H3. We find that the interaction of the peptide with the assembled fragments induces distal conformational rearrangements that suggest an induced fit on a global protein level. A population analysis of an equimolar mixture of an N-terminal and three C-terminal fragments with various affinities for the target peptide revealed predominant assembly of the weakest peptide binder. However, adding a target peptide to this mixture altered the population of the protein complexes such that the combination with the highest affinity for the peptide increased and becomes predominant when adding excess of peptide, highlighting the feasibility of peptide-induced enrichment of best binders from inter-modular fragment mixtures.

Substrate Affinity and Specificity of the ScSth1p Bromodomain Are Fine-Tuned for Versatile Histone Recognition

SummaryBromodomains recognize a wide range of acetylated lysine residues in histones and other nuclear proteins. Substrate specificity is critical for their biological function and arises from unique acetyl-lysine binding sites formed by variable loop regions. Here, we analyzed substrate affinity and specificity of the yeast ScSth1p bromodomain, an essential component of the “Remodels the Structure of Chromatin” complex, and found that the wild-type bromodomain preferentially recognizes H3K14ac and H4K20ac peptides. Mutagenesis studies—guided by our crystal structure determined at 2.7 Å resolution—revealed loop residues Ser1276 and Trp1338 as key determinants for such interactions. Strikingly, point mutations of each of these residues substantially increased peptide binding affinity and selectivity, respectively. Our data demonstrate that the ScSth1p bromodomain is not optimized for binding to an individual acetylation mark, but fine-tuned for interactions with several such modifications, consistent with the versatile and multivalent nature of histone recognition by reader modules such as bromodomains.HighlightsThe ScSth1p bromodomain preferentially recognizes H3K14ac and H4K20ac peptidesSer1276 and Trp1338 are key determinants of substrate affinity and specificityMutations of these residues drastically increase substrate affinity and specificityThe ScSth1p bromodomain is fine-tuned for promiscuous histone tail recognitionGraphical Abstract

Towards a Gamete Matching Platform: Using Immunogenetics and Artificial Intelligence to Predict Recurrent Miscarriage

The degree of Allele sharing of the Human Leukocyte Antigen (HLA) genes has been linked with recurrent miscarriage (RM). However, no clear genetic markers of RM have yet been identified, possibly because of the complexity of interactions between paternal and maternal genes. We propose a methodology to analyse HLA haplotypes from couples either with histories of successful pregnancies or RM. This article describes, for the first time, a method of RM genetic-risk calculation. Novel HLA representation techniques allowed us to create an algorithm (IMMATCH) to retrospectively predict RM with an AUC = 0.71 (p = 0.0035) thanks to high-resolution typing and the use of linear algebra on peptide binding affinity data. The algorithm features an adjustable threshold to increase either sensitivity or specificity. Combining immunogenetics with artificial intelligence could create personalized tools to better understand the genetic causes of unexplained infertility and a gamete matching platform that could increase pregnancy success rates.

Machine Learning-Enhanced T Cell Neoepitope Discovery for Immunotherapy Design

Immune responses mediated by T cells are aimed at specific peptides, designated T cell epitopes, that are recognized when bound to human leukocyte antigen (HLA) molecules. The HLA genes are remarkably polymorphic in the human population allowing a broad and fine-tuned capacity to bind a wide array of peptide sequences. Polymorphisms might generate neoepitopes by impacting the HLA-peptide interaction and potentially alter the level and type of generated T cell responses. Multiple algorithms and tools based on machine learning (ML) have been implemented and are able to predict HLA-peptide binding affinity with considerable accuracy. Challenges in this field include the availability of adequate epitope datasets for training and benchmarking and the development of fully integrated pipelines going from next-generation sequencing to neoepitope prediction and quality analysis metrics. Effectively predicting neoepitopes from in silico data is a demanding task that has been facilitated by ML and will be of great value for the future of personalized immunotherapies against cancer and other diseases.