Untangling Gene Networks
A new grant paves the way for AI research in gene regulatory networks
block
The intricate way that blueprints map out all aspects of a building. The way a family tree maps out a person’s ancestry. The way a circuit board maps the path through its wiring and switches to create electricity.
They all are maps that display complex relationships. UIC College of Engineering researchers have developed AI tools to create their own map of sorts to identify how transcription factors—proteins that regulate gene expression to turn genes on or off—control target genes within individual cells.
Using a grant from NVIDIA, the current leading AI hardware and software technology company, faculty within the Richard and Loan Hill Department of Biomedical Engineering will extend their work to use large language models (LLMs) to integrate scientific literature into the AI tools.
Inspired by collaboration
Last year, Mehrdad Zandigohar, a PhD student in BME professor and associate director of graduate studies Yang Dai’s lab, created scRegulate, an AI tool that takes massive amounts of genomic data and turns them into clear, understandable insights. More specifically, scRegulate infers how genes are regulated in single cells.
The tool avoids the so-called “black box” problem of many AI models that come to conclusions they can’t explain. “The challenge was to make powerful AI models meaningful for biology,” Zandigohar said. “Black-box results are hard to act on, so I set out to design a model whose internal parts map real biology.”
After a computational evaluation and benchmarking, Yang and Zandigohar demonstrated that his tool outperforms existing tools. But while the tool helps them to develop mechanistic insights and select the most relevant transcription factors for experimental validation, it still requires extensive literature reviews and expert knowledge from humans, which remain laborious and sometimes prohibitive.
New innovations
To better integrate scientific literature into the tool, Dai applied for and received the 2025 NVIDIA Academic Grant Program Award, which helps to advance academic research by providing world-class computing access and resources to researchers. Through the grant, Dai and her team were given the NVIDIA DGX Spark for AI, a personal AI supercomputer, which will also allow other UIC researchers to explore AI research.
Access to the AI supercomputer will allow them to develop new ways to use large language models (LLMs) to connect scientific predictions with what’s already known in the literature to create explanations about the gene regulatory networks that other scientists can use.
“The graphics processing unit (GPU)-based tools help us take advantage of exploring differnt large language models,” said Dai, who is also the interim director of the Center for Bioinformatics and Quantitative Biology. “Our lab doesn’t have such a capacity, so this supercomputer allows us that usage. The goal is to try to expand the utility of these tools, through the hardware, the software, and their integration provided by the computer.”
Dai and her team are excited because she has been working in the machine learning area for a long time, but like everyone else, she has felt the challenges of AI. She also noted that without the NVIDIA supercomputer, her research wouldn’t be possible.
To streamline the process of using LLMs, Dai and Zandigohar created RAGulate, a retrieval-augmented generation framework that integrates transcription factor predictions with experimental datasets and biomedical literature. The framework also generates relevant questions for LLMs to produce the most accurate description about the transcription factors and target genes in question in a specific context. Built on NVIDIA DGX Spark’s GPU-accelerated embedding and reranking tools, the team demonstrated that RAGulate improves the biological plausibility of LLM answers and facilitates the prioritization of regulators and their targets for experimental validation.
With the NVIDIA AI supercomputer, Dai’s team aims to accelerate biological discovery by creating an automated system that improves the reliability of LLM responses and reduces false information.
Extending opportunities
Part of the motivation of this research was from a collaboration with other biomedical researchers’ wet labs. Dai is also collaborating with UIC Department of Biochemistry and Molecular Genetics Benjamin Goldberg Professor and Head Jalees Rehman, who is a co-author on the recently published scRegulate journal article. The article was published in Bioinformatics in December of 2025.
“We are currently studying the molecular mechanisms underlying uterine fibroids and hernias using single nucleus multi-omics approaches,” Dai said. “By combining our computational prediction model with a decision-supporting system like RAGulate, we hope to identify key drivers of disease progression and ultimately contribute to therapeutic target discovery.”
These projects are supported by the three funded R01 grants jointly with Dr. Serdar Bulun, Northwestern University Department of Obstetrics and Gynecology Chair and John J. Sciarra Professor of Obstetrics and Gynecology, and his team.
Dai’s broader goal is to expand these AI-powered tools to investigate skin wound healing, including identifying therapeutic targets to enhance tissue repair in diabetic patients, who often experience impaired healing. The NVIDIA AI supercomputer will continue to advance AI-driven biomedical engineering research and strengthen research capacity at UIC.