Our research focuses on three closely related areas in bioinformatics and biomedical informatics as outlined below.
Recent advances in spatial molecular imaging have opened up new possibilities for examining the spatial landscapes and transcriptional profiles of tissues at subcellular resolution, providing deep insights into the spatial microenvironment. The current commercially available technologies for spatial high-plex profiling allow accurate capture of the locations of targeted transcripts, cell locations, and cell boundaries, accompanied by multi-channel immunohistochemistry images. However, interrogating such multi-modality spatial profiling poses challenges that overwhelm traditional machine learning approaches. Advanced ML/AI algorithms are needed to effectively decipher the intricate spatial imaging profiles.
We have demonstrated the efficacy of Graph-based AI algorithms in various spatial profiling studies. One of our novel methods, Single-cell spatial elucidation through image-augmented Graph transformer (SiGra), is designed to reveal spatial domains and enhance the substantially sparse and noisy data in lung cancer spatial profiles. We also proposed a unique adaptive graph model with attention mechanisms, spaCI, which incorporates both spatial locations and gene expression profiles of cells to identify active Ligand-Receptor (L-R) signaling axes across neighboring cells. spaCI has proven effective in revealing L-R interactions in different cancer types, which could otherwise remain hidden due to data sparsity. Furthermore, we developed another novel graph-based convolutional network model, DSTG, to reliably and accurately decompose cell mixtures in spatially resolved transcriptomics spots of pancreatic cancer tissues. These Graph-based AI methods represent a significant advancement in the interpretation of complex spatial omics data, unlocking its potential for various applications in diverse biological research areas.
These stuides have been published in high-impact journals including Nature Communications (IF: 17.69), Briefings In Bioinformatics (IF: 9.5), Genomics, Proteomics & Bioinformatics (IF: 9.5), etc.
Single-cell omics has been increasingly popular, revealing new biological knowledge, and reshaping our understanding of cellular heterogeneity. We have worked on novel methods on single-cell omics data that contribute to the understanding of the interplay of different molecule layers in the tumor microenvironment. For example, we have developed a robust graph-based convolutional network model, scGCN, to achieve effective knowledge transfer across single-cell omics datasets from different sources (e.g., different tissues, platforms, species, and molecular layers). We also developed a joint statistical method for the integrative analysis of single-cell multi-omics data, which enables the detection of coherent regulatory programs and target genes from unpaired single-cell RNA-seq (scRNA-seq) and single-cell ATAC-seq data of mixed-phenotype acute leukemia. We proposed a novel Dual-Stream Subspace Clustering Network model, DS-SCNet, to identify biologically meaningful gene clusters from single-cell RNA-seq data, as well as the other scLM method, to reveal the co-expressed gene signals across single-cell datasets from different sources. Another integrative approach of bioinformatics and multi-scale modeling on scRNA-seq and proteomics data enables the temporal drug response prediction of lung tumor cells.
These studies provide fundamental work for gaining unbiased biological insights from the large single-cell omics data repository, which have been published in Nature Communications (IF: 17.69), Briefings In Bioinformatics (IF: 9.5), etc.
We have been actively working on collaborative multi-omics studies via data-driven informatics approaches to facilitate the understanding of biological mechanisms. For example, we revealed inflammatory Th17 cells presenting distinct transcriptional and metabolic features that may allow precise therapeutic targeting in autoimmunity and infection. In malignant pleural effusion (MPE), we applied single-cell RNA-seq and revealed that intrapleural nano-immunotherapy reprogramed myeloid cells in the tumor microenvironment, reduced MPE and inhibited tumor growth, not only in the pleural cavity but also in the lung parenchyma. For melanoma patients with anti-PD-1 therapy, we used an integrative analysis of bioenergetics, metabolomic profiles, and single-cell RNA-seq data and identified a glycolytic signature characterizing checkpoint inhibitor responders.
Our stuides have been published in high-impact journals including Journal of Experimental Medicine (IF: 17.6), Nature nanotechnology(IF: 40.5), Clinical Cancer Research (IF: 13.8), etc.
With the cancer research community producing vast quantities of molecular and phenotypic data to investigate tumor progression, the concept of big data in cancer field has emerged due to the rapid data accumulation. This presents both opportunities and challenges for discovering novel insights into fundamental questions. Various data types, including genomics, proteomics, transcriptomics, and clinical information such as electronic health records, are available from patient or preclinical samples. It is crucial to analyze such diverse multi-modality data to reveal interpatient heterogeneity and push precision oncology toward clinical care.
Specifically, in the comprehensive molecular study of gynecologic and breast cancers, we identified the clinically significant patient subtypes and potential therapeutic targets, based on multi-modality data including clinical data, copy number alterations, mutations, DNA methylation, and mRNA expression. For patients with brain metastasis, we identified two distinctive prognostic subtypes through an integrative analysis of their transcriptomic, proteomic, and metabolomic molecular profiles. We also revealed the key pathways and metabolites potentially contribute to the observed survival differences between the two subtypes. During the COVID-19 pandemic, we have been working on characterizing the cancer patients within NCATS’ National COVID Cohort Collaborative (N3C), to investigate the risk factors of COVID-19 breakthrough infections and severe outcomes. Our study highlighted patients who received recent immunotherapies or targeted therapies did not have higher risk of overall mortality or breakthrough infection risks, and provided evidence on the effects of COVID-19 on cancer outcomes and the ability to continue specific cancer treatments. This work and our other collaborative studies revealed fundamental insights valuable to the precise protection of cancer population, which has been covered by national news networks such as U.S. News and Fox News.
Our work have been published in high-impact journals including Cancer cell (IF: 35.6), Journal of Clinical Oncology (IF:45.3), etc.
We have established fruitful collaborations across different research disciplines, and have been actively working with scientists across the country on national informatics endeavors for battling cancer. Specifically, we have: 1) Participated in The Cancer Genome Atlas (TCGA) PanCanAtlas Group, where I worked with scientists across the country to comprehensively analyze tumor samples from breast and gynecologic cancers. 2) Contributed to genomics data assembling and sharing at the Wake Forest University site for the AACR Project GENIE (Genomics Evidence Neoplasia Information Exchange). 3) Served as a key member of the NCATS’ National COVID Cohort Collaborative (N3C) oncology domain team. Our collaborative studies (PMID: 35286152; PMID: 34255046; PMID: 34085538) have been covered by national news networks such as U.S. News and Fox News. 4) Served as the key bioinformatician at the Stand Up To Cancer (SU2C) Meg Vosburg T-Cell Lymphoma Dream Team, “Tailoring CAR-based Immunotherapy Strategies to T-cell Lymphoma”, with the goal of developing effective treatment with durable responses that can be applied to a wide variety of T-cell lymphoma malignancies. 5) Joined in the Beat Cancer Childhood (BCC) Research Consortium, aiming to identify new cures for high-risk pediatric cancers through closely collaborating with investigators from diverse backgrounds.