Predictomes: A classifier-curated database of AlphaFold-modeled protein-protein interactions

Abstract

Protein-protein interactions (PPIs) are ubiquitous in biology, yet a comprehensive structural characterization of the PPIs underlying biochemical processes is lacking. Although AlphaFold-Multimer (AF-M) has the potential to fill this knowledge gap, standard AF-M confidence metrics do not reliably separate relevant PPIs from an abundance of false positive predictions. To address this limitation, we used machine learning on well curated datasets to train a Structure Prediction and Omics informed Classifier called SPOC that shows excellent performance in separating true and false PPIs, including in proteome-wide screens. We applied SPOC to an all-by-all matrix of nearly 300 human genome maintenance proteins, generating ~40,000 predictions that can be viewed at predictomes.org, where users can also score their own predictions with SPOC. High confidence PPIs discovered using our approach suggest novel hypotheses in genome maintenance. Our results provide a framework for interpreting large scale AF-M screens and help lay the foundation for a proteome-wide structural interactome.

Delving into human α1,4-galactosyltransferase acceptor specificity: the role of enzyme dimerization

“…To address this, we have incorporated a machine learning method that enhances prediction accuracy by integrating these scores with a comprehensive set of omics and structural data. This integrated approach is encapsulated in the Structure Prediction and Omics-based Classifier (SPOC), specifically designed to evaluate the validity of predicted protein interactions more effectively [Schmid EW et al 2024]. A SPOC score of 0.3 or higher is considered indicative of a high likelihood that the predicted interaction is biologically meaningful and spurious.…”

“…Structure Prediction and Omics-based Classifier (SPOC) is a machine learning tool that integrates structural predictions with omics data to assess the reliability of predicted proteinprotein interactions [Schmid EW et al 2024]. It is valuable for distinguishing true interactions from false positives, offering a robust metric that enhances confidence in computational biology studies.…”

In silicoscreening identifies SHPRH as a novel nucleosome acidic patch interactor

“…Each of the 7,608 proteins was "folded" with an H. sapiens H2A-H2B dimer in ColabFold, and each trimeric protein combination was predicted using 3 of 5 Al-phaFold-Multimer models, yielding a total of 22,824 predicted structures (Methods). Although several analysis methods have been developed to rank predictions from large scale AlphaFold multimer screens, such as pDOCKQ2 10 and SPOC score 11 , these methods are not suitable for identifying specific types of interactions such as those that occur during acidic patch binding. We therefore created a novel analysis pipeline specifically designed to examine residue interactions required for acidic patch binding.…”