How the human mitochondrial interaction dataset was built
The Human MitoMatch dataset was created by repurposing the machine-learning model AlphaFold-Multimer to predict protein-protein interactions across the entire human mitochondrial proteome. We screened all 630,003 pairwise combinations amongst the 1,123 human mitochondrial proteins. Defining an ipTM cutoff using a relevant mitochondrial benchmark dataset, we identified 2,895 positive interactions as high-confidence hits. For each hit we provide multiple confidence metrics along with a mapping of validation through experimental structures or literature curation.
This effort was led by Gohil Lab at the Texas A&M University. The dataset can be downloaded through the website or from Zenodo. Please refer the FAQ or contact us if you have any questions. Please cite our paper if you make use of the MitoMatch dataset.