Quantitative machine learning of protein interactions reveals the multiscale organization and molecular syntax of signaling networks

Cells utilize intricate networks of short-lived protein-protein interactions, facilitated by modular peptide-binding domains and unstructured peptidic motifs, for precise information processing. The manner in which these networks execute computations through intricate intra- and intermolecular mechanisms remains largely unknown due to the limited and non-quantitative nature of existing human interactome maps.

In this study, researchers introduce a quantitative statistical mechanical modeling (QSM) approach designed for machine learning domain-peptide affinities with experimental-level accuracy. Utilizing a novel data harmonization algorithm and a biophysically informed neural network architecture, QSM is capable of predicting dissociation constants directly from amino acid sequences, complete with calibrated confidence.

Applying QSM to construct comprehensive drafts of human signaling networks, the researchers delve into these networks across three distinct physical scales: recognition mechanisms of modular binding domains, combinatorial logic of multi-dentate proteins, and pathways inferred through de novo inference of protein interaction networks.

The findings reveal that (i) modular domains, characterized by their binding preferences, selectivities, and strengths, can be grouped into a limited number of biophysical equivalence groups; (ii) these domains, in conjunction with peptidic motifs, are syntactically combined within proteins to generate multivalent recognition mechanisms; and (iii) the organization of cellular function can be traced back to algorithmically discernible modules, originating from domain-mediated interactions.

In summary, these analyses present a practical roadmap towards a thorough, mechanistic, and simulatable understanding of the systems biology underpinning cellular signaling processes.