PROJECT SUMMARY/ABSTRACT: Overcoming cancer drug resistance is a major unmet clinical need across many cancer types, including difficult-to-treat diseases such as gastroesophageal (GEA) and ovarian cancer. In principle, it should be possible to prioritize therapeutic regimen(s) given the molecular and cellular classification of such tumors. However, two bottlenecks prevent sustained progress. First, biopsies from refractory disease are challenging to routinely obtain and evaluate, both for clinical and research purposes. Second, even when such material is procured, it remains challenging to accurately predict therapeutic sensitivities from baseline tumor and immune profiling. Further, intra-tumoral heterogeneity can render distinct biopsies taken from the same patient tumor discordant. In this revised IMAT proposal we capitalize on the clinical observation that upwards of 40% of GEA and 70% of ovarian patients develop peritoneal carcinomatosis and ascites, a plentiful, heterogenous and molecularly representative sample format that is easily accessible since routine drains occur during patient care. However, in standard practice, such drains are discarded with little detailed evaluation or analysis - and given highly variable tumor purities, bulk profiling is rarely informative. To address these challenges, we have developed the first generation of AscitesPredict, a zero passage ‘ex vivo tumor biosensor’ technology that uses single cell technologies to evaluate cell identities and therapeutic sensitivities during a 5 day period in which viability is preserved. AscitesPredict utilizes high dimensional label-free single cell image-based morphological profiling and machine learning applied to brightfield microscopy which can be collected continuously. Thus far, we have developed AscitesPredict by profiling over 35 samples, made initial assessments of technological reproducibility, and have optimized initial workflows. In this revised proposal we take the next steps to h