Linking signaling to gene expression with mechanistic modeling and machine learning

The National Institute of Allergy and Infectious Diseases awarded the University of Pittsburgh $100,228 on February 18, 2027, under the Allergy and Infectious Diseases Research program (CFDA 93.855) to develop mechanistic models and machine learning approaches that predict how STAT signaling dynamics and promoter sequence features together determine gene-specific inflammatory responses.

The funded project, titled "Linking Signaling to Gene Expression with Mechanistic Modeling and Machine Learning," will develop a novel sequence-to-function neural network that predicts gene expression and STAT DNA binding from promoter sequence and mechanistic model simulations, applied to mouse and human macrophage datasets. The work will identify promoter sequence features that predict STAT binding and gene expression patterns and experimentally validate the functional role of transcription factors associated with discovered sequence motifs using CRISPRi perturbations. The research addresses how STAT1 and STAT3 signaling specificity is achieved and how genes respond differentially to signaling inputs including JAK inhibition.

Performance occurs at Pittsburgh, Pennsylvania. The period of performance extends from the award date through August 31, 2029. This is a project grant, a standard assistance type under NIAID's portfolio supporting investigator-initiated biomedical research in infectious diseases and immunology at nonprofit research institutions.

Abstract

Signal transducer and activator of transcription (stat) proteins link cytokine signaling to gene expression to regulate inflammation. IT remains unclear how the specificity and diversity of stat-mediated gene expression is achieved mechanistically. While the canonical JAK-stat signaling pathway is well characterized, to our knowledge, no other studies have systematically examined how promoter sequence features interact with dynamic stat signaling to decode cytokine-specific transcriptional outcomes. This GAP persists because prior studies analyze signaling pathways or sequence features in isolation, rarely integrating both perspectives within a unified framework. Our lab recently published a computational pipeline combining a mechanistic model of cytokine-induced stat signaling dynamics with machine learning predictions of gene expression (cheemalavagu ET AL. 2024). Although effective at predicting genes with varied sensitivity to JAK2 inhibition, this approach cannot provided mechanistic insight into why specific genes respond differently to signaling inputs. Our central hypothesis is that promoter sequence features and cooperating transcription factor motifs can predict how dynamic STAT1 and STAT3 signaling is translated into stimulus- and gene-specific expression patterns, including response to JAK inhibition. Aim 1 develops a novel sequence-to-function (S2F) neural network that predicts gene expression and stat DNA binding from promoter sequence and mechanistic model simulated stat trajectories, applied to both mouse and human macrophage datasets. Aim 2 identifies promoter sequence features that predict stat binding and gene expression and experimentally validates the functional role of transcription factors associated with these sequence motifs using crispri perturbations. Successful completion of this project will define how promoter sequence shapes sensitivity to cytokine and JAK2 inhibition while also identifying novel mechanisms regulating the gene-level decoding of cytokines signals, guiding the development of more targeted therapies that shape immune function.