Misclassified: identification of zoonotic transition biomarker candidates for influenza A viruses using deep neural network

Introduction: Zoonotic transition of Influenza A viruses is the cause of epidemics with high rates of morbidity and mortality. Predicting which viral strains are likely to transition from their genetic sequence could help in the prevention and response against these zoonotic strains. We hypothesized that features predictive of viral hosts could be leveraged to identify biomarkers of zoonotic viral transition. Methods: We trained deep learning models to predict viral hosts based on the virus mRNA or protein sequences. Our multi-host dataset contained 848,630 unique nucleotide sequences obtained from the NCBI Influenza Virus and Influenza Research Databases. Each sequence, representing one gene from one viral strain, was classified into one of the three host categories: Avian, Human, and Swine. Trained models were analyzed using various neural network interpretation methods to identify interesting candidates for zoonotic transition biomarkers. Results: Using mRNA sequences as input led to higher prediction accuracies than amino acids, suggesting that the codon sequence contains information relevant to viral hosts that is lost during protein translation. UMAP visualization of the latent space of our classifiers showed that viral sequences clustered according to their host of origin. Interestingly, sequences from pandemic zoonotic viral strains localized at the margins between hosts, while zoonotic sequences incapable of Human-to-Human transmission localized with non-zoonotic viruses from the same host. In addition, host prediction for pandemic zoonotic sequences had low prediction accuracy, which was not the case for the other zoonotic strains. This supports our hypothesis that ambiguously predicted viral sequences bear features associated with cross-species infectivity. Finally, we compared misclassified sequences to well-classified ones to extract interesting candidates for zoonotic transition biomarkers. While features varied significantly between pairs of species and viral genes, several codons were conserved in Swine-to-Human and Avian-to-Human misclassified sequences, and in particular in the NA, HA, and NP genes, suggesting their importance for zoonosis in Humans. Discussion: Analysis of viral sequences using neural network interpretation approaches revealed important genetic differences between zoonotic viruses with pandemic potential, compared to non-zoonotic viral strains or zoonotic viruses incapable of Human-to-Human transmission.

CAMAP: Artificial neural networks unveil the role of codon arrangement in modulating MHC-I peptides presentation

MHC-I associated peptides (MAPs) play a central role in the elimination of virus-infected and neoplastic cells by CD8 T cells. However, accurately predicting the MAP repertoire remains difficult, because only a fraction of the transcriptome generates MAPs. In this study, we investigated whether codon arrangement (usage and placement) regulates MAP biogenesis. We developed an artificial neural network called Codon Arrangement MAP Predictor (CAMAP), predicting MAP presentation solely from mRNA sequences flanking the MAP-coding codons (MCCs), while excluding the MCC per se. CAMAP predictions were significantly more accurate when using original codon sequences than shuffled codon sequences which reflect amino acid usage. Furthermore, predictions were independent of mRNA expression and MAP binding affinity to MHC-I molecules and applied to several cell types and species. Combining MAP ligand scores, transcript expression level and CAMAP scores was particularly useful to increase MAP prediction accuracy. Using an in vitro assay, we showed that varying the synonymous codons in the regions flanking the MCCs (without changing the amino acid sequence) resulted in significant modulation of MAP presentation at the cell surface. Taken together, our results demonstrate the role of codon arrangement in the regulation of MAP presentation and support integration of both translational and post-translational events in predictive algorithms to ameliorate modeling of the immunopeptidome.

MHC-I associated peptides (MAPs) are small fragments of intracellular proteins presented at the surface of cells and used by the immune system to detect and eliminate cancerous or virus-infected cells. While it is theoretically possible to predict which portions of the intracellular proteins will be naturally processed by the cells to ultimately reach the surface, current methodologies have prohibitively high false discovery rates. Here we introduce an artificial neural network called Codon Arrangement MAP Predictor (CAMAP) which integrates information from mRNA-to-protein translation to other factors regulating MAP biogenesis (e.g. MAP ligand score and transcript expression levels) to improve MAP prediction accuracy. While most MAP predictive approaches focus on MAP sequences per se, CAMAP’s novelty is to analyze the MAP-flanking mRNA sequences, thereby providing completely independent information for MAP prediction. We show on several datasets that the integration of CAMAP scores with other known factors involved in MAP presentation (i.e. MAP ligand score and mRNA expression) significantly improves MAP prediction accuracy, and further validate CAMAP learned features using an in-vitro assay. These findings may have major implications for the design of vaccines against cancers and viruses, and in times of pandemics could accelerate the identification of relevant MAPs of viral origins.

MAPDP: A Cloud-Based Computational Platform for Immunopeptidomics Analyses

The immunopeptidome corresponds to the repertoire of peptides presented at the cell surface by the major histocompatibility complex (MHC) molecules. Cytotoxic T cells scan this repertoire to identify nonself antigens that can arise from tumors or infected cells. The identification of actionable antigenic targets is key to the development of therapeutic cancer vaccines, T-cell therapy, and other T-cell receptor-based biologics. The growing clinical interest for immunopeptidomics has accelerated the development of high throughput proteogenomic platforms that provide a system-level analysis of MHC-associated peptides. Improvement in sensitivity and throughput of mass spectrometers now allows the detection of a few thousands of peptides from less than 100 million cells. To manage the amount of data generated by these instruments, we have developed the MHC-associated peptide discovery platform (MAPDP), a novel open-source cloud-based computational platform for immunopeptidomic analyses. It provides convenient access from a web portal to immunopeptidomes stored in the database, filtering tools, various visualizations, annotations (e.g., IEDB, dbSNP, gnomAD), peptide-binding affinity prediction (mhcflurry, NetMHC), HLA genotyping, and the generation of personalized proteome databases. MAPDP functionalities are demonstrated here by the discovery of MHC peptides featuring new genetic variants identified in two previously published ovarian carcinoma data sets.

Qualitative Changes in Cortical Thymic Epithelial Cells Drive Postpartum Thymic Regeneration

During gestation, sex hormones cause a significant thymic involution which enhances fertility. This thymic involution is rapidly corrected following parturition. As thymic epithelial cells (TECs) are responsible for the regulation of thymopoiesis, we analyzed the sequential phenotypic and transcriptomic changes in TECs during the postpartum period in order to identify mechanisms triggering postpartum thymic regeneration. In particular, we performed flow cytometry analyses and deep RNA-sequencing on purified TEC subsets at several time points before and after parturition. We report that pregnancy-induced involution is not caused by loss of TECs since their number does not change during or after pregnancy. However, during pregnancy, we observed a significant depletion of all thymocyte subsets downstream of the double-negative 1 (DN1) differentiation stage. Variations in thymocyte numbers correlated with conspicuous changes in the transcriptome of cortical TECs (cTECs). The transcriptomic changes affected predominantly cTEC expression of Foxn1, its targets and several genes that are essential for thymopoiesis. By contrast, medullary TECs (mTECs) showed very little transcriptomic changes in the early postpartum regenerative phase, but seemed to respond to the expansion of single-positive (SP) thymocytes in the late phase of regeneration. Together, these results show that postpartum thymic regeneration is orchestrated by variations in expression of a well-defined subset of cTEC genes, that occur very early after parturition.

Detection of Quiescent Radioresistant Epithelial Progenitors in the Adult Thymus

Thymic aging precedes that of other organs and is initiated by the gradual loss of thymic epithelial cells (TECs). Based on in vitro culture and transplantation assays, recent studies have reported on the presence of thymic epithelial progenitor cells (TEPCs) in young adult mice. However, the physiological role and properties of TEPC populations reported to date remain unclear. Using an in vivo label-retention assay, we previously identified a population of quiescent but non-senescent TECs. The goals of this study were therefore (i) to evaluate the contribution of these quiescent TECs to thymic regeneration following irradiation-induced acute thymic injury and (ii) to characterize their phenotypic and molecular profiles using flow cytometry, immunohistology, and transcriptome sequencing. We report that while UEA1⁺ cells cycle the most in steady state, they are greatly affected by irradiation, leading to cell loss and proliferative arrest following acute thymic involution. On the opposite, the UEA1^– subset of quiescent TECs is radioresistant and proliferate in situ following acute thymic involution, thereby contributing to thymic regeneration in 28- to 30-week-old mice. UEA1^– quiescent TECs display an undifferentiated phenotype (co-expression of K8 and K5 cytokeratins) and express high levels of genes that regulate stem cell activity in different tissues (e.g., Podxl and Ptprz1). In addition, two features suggest that UEA1^– quiescent TECs occupy discrete stromal niches: (i) their preferential location in clusters adjacent to the cortico-medullary junction and (ii) their high expression of genes involved in cross talk with mesenchymal cells. The ability of UEA1^– quiescent TECs to participate to TEC regeneration qualifies them as in vivo progenitor cells particularly relevant in the context of regeneration following acute thymic injury.

MHC class I-associated peptides derive from selective regions of the human genome

MHC class I-associated peptides (MAPs) define the immune self for CD8+ T lymphocytes and are key targets of cancer immunosurveillance. Here, the goals of our work were to determine whether the entire set of protein-coding genes could generate MAPs and whether specific features influence the ability of discrete genes to generate MAPs. Using proteogenomics, we have identified 25,270 MAPs isolated from the B lymphocytes of 18 individuals who collectively expressed 27 high-frequency HLA-A,B allotypes. The entire MAP repertoire presented by these 27 allotypes covered only 10% of the exomic sequences expressed in B lymphocytes. Indeed, 41% of expressed protein-coding genes generated no MAPs, while 59% of genes generated up to 64 MAPs, often derived from adjacent regions and presented by different allotypes. We next identified several features of transcripts and proteins associated with efficient MAP production. From these data, we built a logistic regression model that predicts with good accuracy whether a gene generates MAPs. Our results show preferential selection of MAPs from a limited repertoire of proteins with distinctive features. The notion that the MHC class I immunopeptidome presents only a small fraction of the protein-coding genome for monitoring by the immune system has profound implications in autoimmunity and cancer immunology.

pyGeno: A Python package for precision medicine and proteogenomics

pyGeno is a Python package mainly intended for precision medicine applications that revolve around genomics and proteomics. It integrates reference sequences and annotations from Ensembl, genomic polymorphisms from the dbSNP database and data from next-gen sequencing into an easy to use, memory-efficient and fast framework, therefore allowing the user to easily explore subject-specific genomes and proteomes. Compared to a standalone

program, pyGeno gives the user access to the complete expressivity of Python, a general programming language. Its range of application therefore encompasses both short scripts and large scale genome-wide studies.

pyGeno: A Python package for precision medicine and proteogenomics

pyGeno is a Python package mainly intended for precision medicine applications that revolve around genomics and proteomics. It integrates reference sequences and annotations from Ensembl, genomic polymorphisms from the dbSNP database and data from next-gen sequencing into an easy to use, memory-efficient and fast framework, therefore allowing the user to easily explore subject-specific genomes and proteomes. Compared to a standalone program, pyGeno gives the user access to the complete expressivity of Python, a general programming language. Its range of application therefore encompasses both short scripts and large scale genome-wide studies.