The robot vacuum cleaner Bishal Shrestha built for a ninth grade science project still sits in his home in Kathmandu, Nepal—a reminder of his middle school years, when he learned of the life-enhancing potential of science and discovery.
Today, Shrestha has come a long way since he constructed that automated vacuum. As a University of Miami doctoral student in computer science, he has developed an artificial intelligence tool that detects unusual changes in the 3D structure of DNA, opening potential new pathways that could help researchers uncover new clues to disease.
"We often describe DNA as the blueprint for life, but the role it plays inside a cell involves a lot more than the genetic instructions it carries," Shrestha said. "Think of your DNA as a long piece of string packed inside a tiny cell. The string isn't just sitting there. It folds and organizes itself into complex three-dimensional structures that help determine which genes are turned on or off and how the cell functions."
The tool created by Shrestha and his mentor, Zheng Wang, uses deep learning to identify abnormal patterns in the way DNA interacts and folds inside cells.
"For the past several decades, scientists have tried to understand disease or genetics from a one-dimensional DNA sequence," said Wang, an associate professor who holds appointments in the departments of computer science and biology in the College of Arts and Sciences and at Sylvester Comprehensive Cancer Center.
"Recently, the scientific community realized that the 3D structure of DNA, when it is folded in the nucleus, is very important biologically, as it can bring two sequentially remote segments of the DNA close to each other in 3D space, which has biological significance," Wang continued. "What we did with our tool is to look at that with another dimension added: time. That is, we used AI to check the dynamics of the 3D genome structures over a time."
Shrestha and Wang trained their AI tool, called HiC4D-SPOT, with normal cells over a course of time, allowing it to predict what the 3D structural dynamics of the genome should look like. If the tool encounters a pattern that differs from what it has learned, it flags the difference as a potential anomaly for further investigation.
They tested their tool using simulated abnormalities as well as real biological data, with the tool successfully identifying several types of unusual changes in DNA organization.
Shrestha points out that the published research shows the structural changes flagged by the AI tool correspond to real biological events, such as the reorganization of DNA as stem cells become heart muscle cells and the loss and recovery of DNA loops when a key protein complex is removed and restored. These flagged changes, he said, give scientists a more complete picture of how and when the genome's 3D structure changes, helping them pinpoint the regions that matter.
Their study, published in the journal Briefings in Bioinformatics, provides a promising look at how AI-based anomaly detection can become a useful addition to the tools scientists use to study genome organization.
For Shrestha, the study is a continuation of his work in applying deep learning to decode the 3D genome. He is the lead author on another study of an AI model that predicts gene expression in individual cells using 3D chromosomal interaction data, reducing the need for simultaneous gene expression measurements.
He has also co-authored other studies applying advanced AI to protein-RNA modeling, whole-genome 3D reconstruction, and the discovery of a druggable epigenetic switch linked to nerve regeneration after spinal cord injury.
"When I interviewed Bishal when he was still in Nepal, I could feel that he was a smart, polite, mature young man who had a clear career goal in mind, or simply the motivations of wanting to be successful. Ever since he has arrived at UM, he has always conducted research in a rigorous and passionate way," Wang said.
"Our lab has established incredible momentum in leveraging artificial intelligence to decode the genome," Shrestha said. "Research is where my focus sits. These are things I build for the craft of it, when an idea is worth making real."
