Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn about ATOMICA, a geometric deep learning model that generates universal atomic-scale representations of intermolecular interfaces across diverse biomolecular modalities including small molecules, metal ions, amino acids, and nucleic acids. Discover how this self-supervised approach trains on over 2 million interaction complexes using denoising and masking objectives to create hierarchical embeddings at atomic, chemical block, and molecular interface levels. Explore how the model generalizes across molecular classes and recovers shared physicochemical features without supervision, with its latent space capturing compositional and chemical similarities across interaction types while following scaling laws that improve with increasing biomolecular data modalities. Examine the construction of five modality-specific interfaceome networks called ATOMICANets that connect proteins based on interaction similarity with ions, small molecules, nucleic acids, lipids, and proteins, enabling identification of disease pathways across 27 conditions and prediction of disease-associated proteins in autoimmune neuropathies and lymphoma. Understand how ATOMICA annotates the dark proteome by predicting 2,646 previously uncharacterized ligand-binding sites, including putative zinc finger motifs and transmembrane cytochrome subunits, demonstrating systematic annotation capabilities for molecular interactions across the entire proteome.
