[F31] Interpretable Computational Methods for Unraveling Pathogenic Immune Dysregulation Signals in Single-Cell Omics Data

Description

Project Summary/Abstract Autoimmune diseases such as rheumatoid arthritis (RA) show heterogeneous and uncertain disease trajectories driven by diverse immune cell states, yet mechanisms remain poorly understood. Single-cell sequencing allows immune cell profiling at unprecedented resolution, facilitating the identification of cell states that drive autoimmune pathogenesis. We recently identified expanded T-cell subsets in individuals at risk for RA, suggesting early immune remodeling prior to clinical onset. Existing methods struggle to detect continuous disease-associated transcriptional programs, or model nonlinear, clinically relevant, and temporally dynamic phenotypes. Artificial Intelligence (AI) approaches could address these limitations, however, lack of interpretability prevents their predictions from translating into mechanistic hypotheses. Explainable artificial intelligence (XAI) promises to overcome this barrier by revealing cell-level importance, although its ability to identify continuous transitions, rather than discrete clusters, remains an open question. These computationally derived phenotypes must be experimentally validated to confirm their functional significance, and clinical relevance. Our goal is to develop biologically interpretable computational frameworks that link immune cell states with clinical outcomes to illuminate how transcriptional programs evolve during disease progression. We hypothesize that our novel XAI approaches will isolate the continuous remodeling of T cell populations characterized by gradual shifts in molecular programs that drive loss of tolerance predictive of autoimmune disease progression. Aim 1: We will design a framework that leverages XAI to generate cell- level interpretable scores that describe each cell's contribution to the clinical classification on an RA single-cell dataset to detect T cell populations that predict disease pathogenesis. We will experimentally validate the identified populations using flow cytometry and functional assays to confirm their role in autoimmune pathogenesis. Aim 2: We will develop a time-aware attention-based multiple instance learning framework to identify disease trajectories using our in-house single-cell longitudinal RA datasets and identify a T cell subset whose abundance shifts correlate with RA progression. On patient samples from an external cohort, we will validate the clinical relevance of the predicted cell subsets by correlating the subset abundance with clinical metrics. We anticipate that this work will link predictive model features to interpretable cellular states including a population largely encompassing Tph, Th1, Th17, and Th22 cells and experimentally uncover the molecular features predictive of disease activity. The proposed XAI framework will provide a transparent, biologically grounded approach for discovering candidate biomarkers and therapeutic targets, ultimately advancing precision medicine for autoimmune disorders. I will gain valu Istituzione: UNIVERSITY OF COLORADO DENVER PI: Jade M Young Progetto: 1F31AI203002-01

Sectors

Source

Source: reporter.nih.gov