Mitochondria: AI predicts thousands of protein partners
· BehindLines newsroom
Researchers at Texas A&M AgriLife Research have used an artificial intelligence tool to map the likely partners of mitochondrial proteins. Mitochondria are the structures that supply energy to human cells. According to Phys.org and the AgriLife Today press service, the results appeared in the journal Nature Communications under the title "The predicted interactome of the human mitochondrial proteome."
A prediction tool put to a new use
According to AgriLife Today, about 1,100 proteins are known to exist in human mitochondria, and the function of many of them remains unknown. Mutations affecting nearly 30% of them have been linked to diseases, including cancers, Parkinson's disease, Alzheimer's disease and rare inherited metabolic disorders.
The team turned to AlphaFold-Multimer, a tool Google DeepMind unveiled in 2022 to predict the shapes that proteins form when they bind together. According to AgriLife Today, it screened 630,003 possible pairs. The method was first tried on proteins of the mitochondrial copper pathway. The first author, Abhinav Swaminathan, a researcher in Vishal Gohil's laboratory, says it "worked really well," which led the team to extend it to all of the proteins.
What the map reveals
The screen flagged 2,895 likely interactions, more than half of which had never been reported before, according to AgriLife Today. Phys.org adds that it also gave 85 previously uncharacterized proteins their first probable partners, which offers clues about their roles.
These predictions were followed by about two years of laboratory work, according to AgriLife Today: several of the predicted partnerships were confirmed in yeast cells and human cells, with the participation of researchers from Washington University in St. Louis. The press release does not say how many predictions were tested or what share was confirmed.
An open resource, with clear limits
The predictions are available free of charge online under the name MitoMatch, so that other laboratories can look up the likely partners of the proteins they study.
The authors remain cautious. Knowing what a protein does is necessary before the diseases linked to it can be treated, and interactions predicted by computer must be checked by experiment. Abhinav Swaminathan stresses that the lab is a "traditional" biochemistry and genetics laboratory, not an AI lab, and that treatments for mitochondria-related diseases may still be far off. The team is continuing its tests with other researchers.