Existing functional description of genes are categorical, discrete, and mostly through manual process. In this work, we explore the idea of gene embedding, distributed representation of genes, in the spirit of word embedding. From a pure data-driven fashion, we trained a 200-dimension vector representation of all human genes, using gene co-expression patterns in 984 data sets from the GEO databases. These vectors capture functional relatedness of genes in terms of recovering known pathways - the average inner product (similarity) of genes within a pathway is 1.52X greater than that of random genes. Using t-SNE, we produced a gene co-expression map that shows local concentrations of tissue specific genes. We also illustrated the usefulness of the embedded gene vectors, laden with rich information on gene co-expression patterns, in tasks such as gene-gene interaction prediction. We proposed a machine learning method that utilizes transcriptome-wide gene co-expression to generate a distributed representation of genes. We further demonstrated the utility of our distribution by predicting gene-gene interaction based solely on gene names. The distributed representation of genes could be useful for more bioinformatics applications.

Human complex traits and common diseases show tissue- and cell-type- specificity. Recently, single-cell RNA sequencing (scRNA-seq) technology has successfully depicted cellular heterogeneity in human tissue, providing an unprecedented opportunity to understand the context-specific expression of complex trait-associated genes in human tissue-cell types (TCs). Here, we present the first web-based application to quickly assess the cell-type-specificity of genes, named Web-based Cell-type Specific Enrichment Analysis of Genes (WebCSEA, available at https://bioinfo.uth.edu/webcsea/). Specifically, we curated a total of 111 scRNA-seq panels of human tissues and 1,355 TCs from 61 different general tissues across 11 human organ systems. We adapted our previous decoding tissue-specificity (deTS) algorithm to measure the enrichment for each tissue-cell type (TC). To overcome the potential bias from the number of signature genes between different TCs, we further developed a permutation-based method that accurately estimates the TC-specificity of a given inquiry gene list. WebCSEA also provides an interactive heatmap that displays the cell-type specificity across 1355 human TCs, and other interactive and static visualizations of cell-type specificity by human organ system, developmental stage, and top-ranked tissues and cell types. In short, WebCSEA is a one-click application that provides a comprehensive exploration of the TC-specificity of genes among human major TC map.

Intratumoral infiltration of myeloid-derived suppressor cells (MDSCs) is known to promote neoplastic growth by inhibiting the tumoricidal activity of T cells. However, direct interactions between patient-derived MDSCs and circulating tumors cells (CTCs) within the microenvironment of blood remain unexplored. Dissecting interplays between CTCs and circulatory MDSCs by heterotypic CTC/MDSC clustering is critical as a key mechanism to promote CTC survival and sustain the metastatic process. We characterized CTCs and polymorphonuclear-MDSCs (PMN-MDSCs) isolated in parallel from peripheral blood of metastatic melanoma and breast cancer patients by multi-parametric flow cytometry. Transplantation of both cell populations in the systemic circulation of mice revealed significantly enhanced dissemination and metastasis in mice co-injected with CTCs and PMN-MDSCs compared to mice injected with CTCs or MDSCs alone. Notably, CTC/PMN-MDSC clusters were detected in vitro and in vivo either in patients’ blood or by longitudinal monitoring of blood from animals. This was coupled with in vitro co-culturing of cell populations, demonstrating that CTCs formed physical clusters with PMN-MDSCs; and induced their pro-tumorigenic differentiation through paracrine Nodal signaling, augmenting the production of reactive oxygen species (ROS) by PMN-MDSCs. These findings were validated by detecting significantly higher Nodal and ROS levels in blood of cancer patients in the presence of naïve, heterotypic CTC/PMN-MDSC clusters. Augmented PMN-MDSC ROS upregulated Notch1 receptor expression in CTCs through the ROS-NRF2-ARE axis, thus priming CTCs to respond to ligand-mediated (Jagged1) Notch activation. Jagged1-expressing PMN-MDSCs contributed to enhanced Notch activation in CTCs by engagement of Notch1 receptor. The reciprocity of CTC/PMN-MDSC bi-directional paracrine interactions and signaling was functionally validated in inhibitor-based analyses, demonstrating that combined Nodal and ROS inhibition abrogated CTC/PMN-MDSC interactions and led to a reduction of CTC survival and proliferation. This study provides seminal evidence showing that PMN-MDSCs, additive to their immuno-suppressive roles, directly interact with CTCs and promote their dissemination and metastatic potency. Targeting CTC/PMN-MDSC heterotypic clusters and associated crosstalks can therefore represent a novel therapeutic avenue for limiting hematogenous spread of metastatic disease.

Single-cell RNA sequencing (scRNA-Seq) is rapidly becoming a powerful tool for high-throughput transcriptomic analysis of cell states and dynamics at the single cell level. Both the number and quality of scRNA-Seq datasets have dramatically increased recently. A database that can comprehensively collect, curate, and compare expression features of scRNA-Seq data in humans has not yet been built. Here, we present scRNASeqDB, a database that includes almost all the currently available human single cell transcriptome datasets (n = 38) covering 200 human cell lines or cell types and 13,440 samples. Our online web interface allows users to rank the expression profiles of the genes of interest across different cell types. It also provides tools to query and visualize data, including Gene Ontology and pathway annotations for differentially expressed genes between cell types or groups. The scRNASeqDB is a useful resource for single cell transcriptional studies. This database is publicly available at bioinfo.uth.edu/scrnaseqdb/.

Drug response differs substantially in cancer patients due to inter- and intra-tumor heterogeneity. Particularly, transcriptome context, especially tumor microenvironment, has been shown playing a significant role in shaping the actual treatment outcome. In this study, we develop a deep variational autoencoder (VAE) model to compress thousands of genes into latent vectors in a low-dimensional space. We then demonstrate that these encoded vectors could accurately impute drug response, outperform standard signature-gene based approaches, and appropriately control the overfitting problem. We apply rigorous quality assessment and validation, including assessing the impact of cell line lineage, cross-validation, cross-panel evaluation, and application in independent clinical data sets, to warrant the accuracy of the imputed drug response in both cell lines and cancer samples. Specifically, the expression-regulated component (EReX) of the observed drug response achieves high correlation across panels. Using the well-trained models, we impute drug response of The Cancer Genome Atlas data and investigate the features and signatures associated with the imputed drug response, including cell line origins, somatic mutations and tumor mutation burdens, tumor microenvironment, and confounding factors. In summary, our deep learning method and the results are useful for the study of signatures and markers of drug response.

Mutations in kinases are abundant and critical to study signaling pathways and regulatory roles in human disease, especially in cancer. Somatic mutations in kinase genes can affect drug treatment, both sensitivity and resistance, to clinically used kinase inhibitors. Here, we present a newly constructed database, KinaseMD (kinase mutations and drug response), to structurally and functionally annotate kinase mutations. KinaseMD integrates 679 374 somatic mutations, 251 522 network-rewiring events, and 390 460 drug response records curated from various sources for 547 kinases. We uniquely annotate the mutations and kinase inhibitor response in four types of protein substructures (gatekeeper, A-loop, G-loop and αC-helix) that are linked to kinase inhibitor resistance in literature. In addition, we annotate functional mutations that may rewire kinase regulatory network and report four phosphorylation signals (gain, loss, up-regulation and down-regulation). Overall, KinaseMD provides the most updated information on mutations, unique annotations of drug response especially drug resistance and functional sites of kinases. KinaseMD is accessible at https://bioinfo.uth.edu/kmd/, having functions for searching, browsing and downloading data. To our knowledge, there has been no systematic annotation of these structural mutations linking to kinase inhibitor response. In summary, KinaseMD is a centralized database for kinase mutations and drug response.