Combining old techniques and new methods, Dr. Nault’s new paper lays out a path to assist future research

Combining the old and the new – internet-based public information and groundbreaking Artificial Intelligence (AI)- Dr. Rance Nault and his data scientist, Keji Yuan, have published a paper that seeks to understand how using a graph neural network can help identify disruptions in metabolism resulting from chemical exposures.

The study’s goal was to apply a network-based graph neural network (GNN) to uncover novel or hidden metabolic changes in response to a toxicant (in this case, dioxin).

“We applied AI to try and figure out how toxicants change metabolism,” Dr. Nault said. “Metabolism is a big, connected network of reactions, and when we look at the effects of these chemicals, we look at them as individual pathways, which are treated as silos. What’s really nice with AI is it can capture the connections between the pathways in the form of a graph on which AI can be trained on.”

A GNN model based on the mouse model pathways was trained and validated on 7,689 publicly available transcriptomic samples from 26 mouse tissues.

“We wanted to determine how the metabolism was affected and see if there are things that are missing from the ways we traditionally do it, and this was a way to find what was hidden” Dr. Nault said. “Because the pathways are all connected and we usually examine them independently, there are things in that network that are hidden, which we can now detect.”

That was the AI portion of the research. And as valuable as it was, it was the publicly available information, provided from previously published research, that offered even more insight.

“We hope this will help us make better use of historical data, especially as we try to move away from animal studies,” Dr. Nault said. “There’s a huge amount of data that’s already available on the internet in databases and repositories. So how do we use public information to gain new information? Using AI we wanted to see if we could discover new things that we wouldn’t be able to do in the classical ways.”

As with much research, there is still much to learn. But Dr. Nault sees this as a starting point to expand on the existing scientific knowledge.

“It’s maximizing the value of data that has already been published to help future research,” he said. “The goal is to demonstrate how we can re-use all of this data to gain new knowledge and come up with a more effective direction to follow up.”