Gene Regulation & Disease Genetics

Haohan Wang's group at the University of Illinois Urbana-Champaign develops statistical and deep learning methods to understand how genes are regulated in their tissue context and how genetic variation drives complex diseases, with a long-running focus on Alzheimer's disease. A common thread is building methods whose findings hold beyond a single dataset, by modeling confounders, spatial context, and patient-level heterogeneity.

Key questions

How can gene regulatory networks be inferred from spatial transcriptomics?

The group's SVGRN method combines gene expression, known regulatory interactions, and spatial coordinates in a structural equation model implemented with a conditional variational autoencoder. It refines tissue-level regulation into spot- or cell-specific networks and was presented as an oral talk at ISMB.

Why do Alzheimer's patients differ genetically?

Using a heteroscedastic personalized regression (Het-PR) framework on the ADNI cohort, the group built individualized SNP-effect profiles and found two patient subgroups with divergent performance across five cognitive domains. Variants linked to other neuropsychiatric traits, especially epilepsy, distinguished the subgroups, suggesting shared genetic architecture across brain-related traits.

Selected projects

Spatially Varying Gene Regulation Network Inference from Spatial Transcriptomics (SVGRN)

Li Y, Chen J, Lu T, Tsai N-P, Wang H · Bioinformatics Advances, 2026 · ISMB oral presentation

A deep learning framework that infers spatially resolved, high-resolution gene regulatory networks by conditioning on location and neighborhood information.

Consistently outperforms existing methods on simulated data under diverse settings
Captures spatially varying regulatory programs in seqFISH mouse embryo data and Visium human cutaneous squamous cell carcinoma and fallopian tube data

[Paper] [Code]

Stratifying Alzheimer's Disease by Patient-Specific Genetic Signatures Reveals Cognition-Linked and Cross-Disease Heterogeneity

Zhang J, Yu Z, Zhang X, Wang H · npj Dementia, 2026

Introduces heteroscedastic personalized regression (Het-PR) to build individualized SNP-effect profiles, moving beyond cohort-averaged genetic associations.

Identifies two AD patient subgroups with divergent performance across five cognitive domains in ADNI
Frequently selected variants map to brain-expressed genes; epilepsy-associated variants show the strongest proportional signal, including loci such as SCN1A

[Paper] [Code]

Gene Set Prioritization Guided by Regulatory Networks with p-values through Kernel Mixed Model (KMM)

Wang H, Lopez OL, Wu W, Xing EP · RECOMB; Journal of Computational Biology, 2022

A transcriptome association method that incorporates gene regulatory networks as kernels of a linear mixed model and reports p-values, so that genes are prioritized with statistical guarantees and pathway context. Released as a Python package.

Applied to Alzheimer's disease to identify sets of associated genes
One-line command-line usage, with sparse-matrix support for large networks

[RECOMB paper] [Software paper] [Code]

Talks on this topic

Spatially Varying Cell-specific Gene Regulation Network Inference — NIH Common Fund Data Ecosystem (CFDE) Meeting, Bethesda (Mar. 2025)
Understanding Variations in Regulatory Networks Across Cell Types through Transformer Model with Knowledge on Regulatory Interactions — NIDA, NIH, Bethesda (June 2024)
Deep Learning Methods to Navigate Heterogeneous Data Landscapes for Genetic Insights — Spatial Genomics Seminar, Carl R. Woese Institute for Genomic Biology (Feb. 2024)
Advancing Precision Medicine: Tailored Genomic Insights and AI-Driven Automation in Complex Disease Research — UW–Madison (Feb. 2024); UIUC CS (Mar. 2024)
Understanding Structural Patterns for Early-Diagnosis of Alzheimer's Disease — Stanford AI+Health Seminar (Dec. 2023)
Dealing with Confounding Factors with Deep Neural Networks — Next Generation in Biomedicine Symposium, Broad Institute (Sept. 2019)

Research support

This work is supported by the National Institutes of Health, including a grant on deep learning techniques for cell-type and spatial resolution estimation of regulatory networks.

Invite a talk or collaborate

Haohan Wang gives talks on spatial gene regulation, machine learning for complex-disease genetics, and Alzheimer's disease heterogeneity, and welcomes collaborations with groups generating spatial, single-cell, or genetic cohort data. Contact Haohan Wang at haohanw at illinois.edu.