Deep learning based 3D network biology knowledge portal for precision oncology

/ABSTRACT Interpretation of clinical pathogenesis of somatic mutations is crucial for the advancement of precision cancer medicine, especially given the thousands of cancer genome/exome sequencing data available today and many more to come in the near future. Although traditional linear genomic sequence focused, often “one at a time” approaches have led to seminal discoveries, they now must be complemented by 3D protein structure based methods that take into account rewiring of sub-cellular systems and the molecular “interactome” network during tumor initiation, progression and maintenance. However, 237,787 interactions (94.2% of the current human interactome) do not have any structural information, most of which are not amenable to current structural modeling methods (including AlphaFold-based methods), Here, we propose to develop a big-data-driven deep- learning-based pipeline, named PIONEER (Protein-protein InteractiON IntErfacE pRediction; Aim 1) to generate a comprehensive 3D human interactome with significant improvement in quality and coverage. Importantly, we will integrate PIONEER with the powerful structure-alignment-based PrePPI pipeline to generate atomic- resolution 3D models for the entire human interactome for the first time, addressing a key unmet need in precision oncology and critical for NCI missions. Take advantage of nearly full structural coverage for all individual proteins (by AlphaFold) and protein- protein interactions (PPIs; by PIONEER), we will develop an end-to-end 3D-structurally-informed anisotropic network propagation framework to identify 3D spatial clusters of cancer mutations, especially at PPI interfaces (named “oncoPPIs”), and likely dysregulated network modules/pathways (Aim 2). In Aim 3, we will validate our interface predictions and 3D models using existing large-scale mutagenesis and interactome perturbation data, and through cross-linking mass spectrometry experiments. We will further functionally validate our results using existing cancer proteogenomic and clinical datasets. We have extensive preliminary results confirming that our 3D-structrually-based identification of oncoPPIs and dysregulated modules/pathways significantly correlated with patient survival and treatment across diverse cancer types, providing excellent candidates for developing personalized medicine and treatment strategies, as well as better understanding of molecular mechanisms underlying specific cancer types. Furthermore, we will perform functional validations in cancer cell models and patient-derived tumor organoids through collaborations. We will deploy an interactive web portal to disseminate our results and all of our tools for on-demand 3D model building, network analysis, and cancer multi-omics studies. Our comprehensive multiscale 3D human interactome and the accompanying knowledge portal will greatly reduce the barrier-to-entry for performing systematic structural analysis on a large number of proteins and their interactions, and open the flood gates for such analyses in cancer genetic and genomic studies. Project Number: 1U01CA294154-01A1 | Fiscal Year: 2025 | NIH Institute/Center: National Cancer Institute (NCI) | Principal Investigator: Haiyuan Yu (+1 co-PI) | Institution: CORNELL UNIVERSITY, ITHACA, NY | Award Amount: $413,509 | Activity Code: U01 | Study Section: Special Emphasis Panel[ZCA1 TCRB-9 (J1)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11110063