Predictomes

Predictomes provides a classifier-curated database of AlphaFold-modeled protein-protein interactions that uses machine learning to score interaction likelihood and enables interpretation of large-scale AlphaFold-Multimer screens.

Key Features:

• Structure Prediction and Omics-informed Classifier (SPOC): A machine learning classifier trained on curated datasets that separates true and false AlphaFold-Multimer predictions of protein-protein interactions.
• Proteome-wide Screening Capability: SPOC effectively separates true and false predictions in proteome-wide screens.
• Database of Predictions: Contains approximately 40,000 predictions from an all-by-all matrix of nearly 300 human genome maintenance proteins.
• User Prediction Scoring: Users can score their own AlphaFold-Multimer predictions with SPOC.

Scientific Applications:

• Genome Maintenance Research: High-confidence protein-protein interactions discovered using SPOC enable hypothesis generation in genome maintenance.
• Structural Interactome Construction: Provides a framework for interpreting large-scale AlphaFold-Multimer screens and helps lay the foundation for a proteome-wide structural interactome.

Methodology:

Machine learning was trained on curated datasets to develop the structure prediction and omics-informed classifier (SPOC), which was then applied to an all-by-all matrix of nearly 300 human genome maintenance proteins.