UVA Researchers Develop Powerful AI Tool to Help Unlock Causes of Disease and Speed New Treatments

Scientists at the University of Virginia School of Medicine have developed a new artificial intelligence tool that could help researchers understand why certain genetic mutations cause disease — and potentially shorten the time needed to develop new treatments.

The technology, called ERIS2, uses deep learning to predict how changes in a person’s genetic code affect the stability of proteins inside the body’s cells.

That may sound complicated, but the idea behind the technology is relatively simple.

Proteins perform many of the essential jobs that keep the human body functioning. When a genetic mutation changes a protein’s structure, it can interfere with how it works, sometimes contributing to serious illnesses.

By predicting those changes, ERIS2 could help scientists identify potentially harmful mutations more quickly and focus their research on promising treatments.

A research team led by Nikolay V. Dokholyan, PhD, of UVA’s Department of Neurology developed the tool.

“Understanding how mutations affect protein stability is key to understanding protein function and often disease,” Dokholyan said.

“Predicting how mutations change protein stability helps us understand protein function and its relationship to disease, while also enabling us to design new proteins that can serve as biotechnological, diagnostic, or therapeutic tools.”

Why Protein Stability Matters

To understand the importance of the new technology, it helps to know a little about proteins.

Every cell in the human body contains proteins that carry out essential functions, including repairing damaged tissues, supporting the immune system, and helping cells communicate.

Proteins are made of chains of amino acids that fold into specific three-dimensional shapes.

Those shapes are critical.

A protein’s structure helps determine whether it can perform its intended function. Some proteins act as enzymes that drive chemical reactions, while others transport molecules or help regulate cellular activity.

But a gene mutation can change one of the amino acids used to build a protein.

Even a seemingly small change can sometimes alter how the protein folds or how stable its structure remains.

If a protein becomes unstable, it may not function properly, potentially contributing to inherited genetic disorders, cancer, or other diseases.

Not every mutation is harmful, however. Some have little effect, while others may actually improve a protein’s stability.

Determining which changes matter is one of the challenges facing medical researchers.

Artificial Intelligence Takes on a Difficult Problem

Scientists already have methods for studying how mutations affect proteins, but traditional laboratory experiments can be expensive and time-consuming.

Researchers may need to produce proteins, introduce specific mutations, and measure the effects individually.

The process becomes particularly challenging with complex proteins or large numbers of possible mutations.

ERIS2 is designed to speed that up.

UVA researchers combined experimental protein measurements with detailed structural information and trained their artificial intelligence model on data from more than 100,000 known mutations.

The system learns relationships between changes in a protein’s amino acid sequence, its three-dimensional structure, and the resulting changes in stability.

Researchers can then use the tool to estimate how a new mutation might affect a protein without first conducting every possible laboratory experiment.

The predictions can help identify which mutations deserve closer examination.

Testing Shows Promising Results

The researchers tested ERIS2 on independent datasets kept separate from its training data.

The results, published October 30, 2026, in the scientific journal Structure, showed that the system could predict changes in protein stability with encouraging accuracy.

Across three independent test datasets, the researchers reported correlations between predicted and measured stability changes ranging from 0.51 to 0.69.

Those results suggest the tool can provide useful information, although its predictions are not perfect and still require scientific validation.

That distinction is important.

ERIS2 does not diagnose disease, determine with certainty whether a mutation will cause illness, or replace laboratory research.

Instead, it gives scientists another tool for investigating how genetic changes might influence the body.

By narrowing the possibilities, researchers may spend less time testing unlikely candidates and more time studying mutations and treatments that appear promising.

A Possible Boost for Cancer and Genetic Disease Research

The technology could be especially useful for studying conditions in which protein function plays an important role.

Certain inherited diseases result from mutations that interfere with proteins the body needs to function normally.

Cancer can also develop through genetic changes that affect how cells grow, divide, and respond to signals.

Understanding how mutations alter protein stability may provide clues about what is happening inside affected cells.

That information could eventually help researchers design treatments to correct, stabilize, or otherwise address malfunctioning proteins.

The same technology may also support protein engineering, a field in which scientists design or modify proteins for use in medicine, biotechnology, and diagnostic testing.

Although ERIS2 is a research tool rather than a new medical treatment, its developers believe it could help accelerate discoveries that ultimately benefit patients.

Available Free to Researchers Worldwide

One notable feature of ERIS2 is that UVA is making it available to the scientific community at no cost.

Researchers can access the system online without creating an account or registering.

The tool lets scientists enter information about a protein’s structure and a specific amino acid mutation, then receive a prediction of how that change is likely to affect stability.

The research team hopes that making the system freely available will encourage broader use and further improvements.

Part of UVA’s Growing Biotechnology Effort

The development of ERIS2 aligns with the mission of UVA’s new Paul and Diane Manning Institute of Biotechnology.

The institute focuses on translating scientific discoveries into medical treatments more efficiently, reducing barriers that can delay promising advances from reaching patients.

Artificial intelligence is becoming an increasingly important part of that effort.

By analyzing large quantities of biological information, AI systems can help researchers recognize patterns that might be difficult to detect through traditional methods.

ERIS2 represents another step toward combining computational research with laboratory science to better understand disease.

The research team included Srinivasan Ekambaram, Debarshi Kundu, Swaroop Ghosh, and Dokholyan.

The project received support from the National Science Foundation and Intel, with additional research funding supporting Dokholyan’s work from the National Institutes of Health and NSF.

The researchers reported no financial conflicts of interest.

Turning Information Into Better Treatments

Developing a new medication can take years of research, testing, and clinical trials.

One of the earliest challenges is understanding exactly what goes wrong in the body when a disease develops.

For genetic disorders and some cancers, that often means examining how mutations affect the proteins that cells depend on.

ERIS2 cannot eliminate the need for laboratory experiments or clinical testing.

But it may help researchers answer important questions sooner, giving them a better starting point for designing potential treatments.

For patients waiting for new options, that kind of progress matters.

The promise of ERIS2 isn’t that artificial intelligence has found a cure. It’s that scientists may now have a faster way to understand some of the molecular problems they need to solve before developing a cure.