In contrast to the substantial progress in mapping human genomes, systematic efforts to chart the architecture of human cells are just beginning. I will describe our ongoing efforts to close this gap, including new machine learning approaches and new spatial proteomic datasets that systematically inform the makeup of diseased and normal cells. To construct a comprehensive map of subcellular architecture—including organelles, compartments, and the protein complexes that comprise them—we leverage self-supervised learning techniques to integrate diverse structural and functional omics modalities. This integrative approach yields detailed whole-cell representations of specific human cell types, which are subsequently used to train deep learning models capable of bridging genotype to phenotype. We have applied this framework to key challenges in precision oncology, demonstrating its utility in predicting responses to chemotherapy and CDK4/6 inhibitors, a frontline treatment for advanced breast cancer. The resulting insights into tumor resistance mechanisms inform the rational design of new therapeutic agents and synergistic drug combinations, which we validate through high-throughput screening and patient-derived xenograft models. Building on this foundation, we have collaborated with the broader research community to propose a set of Predictive Oncology Hallmarks—a framework for evaluating the accuracy, robustness, and translational potential of drug response models for clinical deployment.

Link for joining via Microsoft Teams: teams.microsoft.com/l/meetup-join/19%3ameeting_Y2RlYWY3MmEtNjg1Ni00NzJjLTk1MDktYThjZGI5OGM2NmQ2%40thread.v2/0?context=%7b%22Tid%22%3a%22cc95de1b-97f5-4f93-b4ba-fe68b852cf91%22%2c%22Oid%22%3a%22467ab6c0-d72f-4fc9-aec9-3618ee3020ba%22%7d