AI Maps Hidden Protein Partners in Human Mitochondria
Texas A&M researchers used an AI tool to predict more than 2,800 likely protein interactions inside the cell's energy-producing structures.
Updated on Nov. 9, 2026 in Life Sciences
Texas A&M AgriLife Research scientists used an AI tool to predict interactions among human mitochondrial proteins and released the predictions as a free online resource called MitoMatch. The study cataloged 2,895 likely protein interactions, more than half of which had never been reported before.
Why it matters
Mutations in nearly 30% of mitochondrial proteins have been linked to human diseases, and without knowing what these proteins do, researchers struggle to understand what causes those conditions. Traditional lab methods for identifying protein functions are slow, so a faster way to generate candidate partnerships could speed up research across Texas and beyond.
The team screened more than 630,000 protein pairs from roughly 1,100 mitochondrial proteins and cataloged 2,895 likely interactions, over half previously unreported. Only several predicted partnerships were confirmed in laboratory experiments, and the study does not show whether these interactions will yield treatments.
The players
Google DeepMind
Google DeepMind is the Alphabet-owned artificial intelligence research lab that developed the AlphaFold family of protein-structure prediction models.
Texas A&M AgriLife Research
Texas A&M AgriLife Research is the agricultural research arm of Texas A&M University that led the study.
The details
Researchers used AlphaFold-Multimer, an AI tool that predicts the three-dimensional shapes proteins form together, to screen protein pairs. They first tested the approach on mitochondrial copper pathway proteins before applying it to all human mitochondrial proteins, then confirmed several predicted partnerships in yeast and human cells, with some experiments done alongside Washington University School of Medicine in St. Louis.
Timeline
About three years before publication, Swaminathan and Gohil began testing AlphaFold-Multimer.
In 2022, Google DeepMind created and unveiled AlphaFold-Multimer.
Over about two years of lab work, several predicted protein partnerships were confirmed.
On November 9, 2026, the study was published in Nature Communications.
Big Picture
The study extends AlphaFold-Multimer, which Google DeepMind unveiled in 2022, from general structure prediction into systematic mapping of one organelle's protein network. It signals a shift in which AI-predicted interaction screens serve as a first pass before costly wet-lab validation.
For patients and families affected by mitochondrial disease, the MitoMatch resource offers researchers a free starting point for investigating which proteins work together. Any resulting therapies are likely years away, since the study's experimental confirmations covered only several predicted partnerships.
The takeaway
AI-generated protein maps can rapidly narrow the search for disease mechanisms, but predictions still require laboratory confirmation before they guide treatment. Readers following mitochondrial disease research should watch for follow-up validation of the newly catalogued interactions.
Further reading
Explore more Texas Life Sciences coverage for context on research being conducted in the state.
Source note: This article includes information reported by Technology Networks.
