OncoAI - a machine-learning framework to predict cancer cell vulnerability at the gene-regulatory level

The vastly increasing amounts of publicly available single-cell data that are
being obtained from tumor biopsies are posing significant computational
challenges to integrate data across "omics layers", but also to compare
patient samples, cancer types, and model systems. Recent progress in
artificial intelligence provides exciting opportunities to process millions of
single cells, and to derive comprehensive models and key regulatory drivers
of cancer cell states. Here, we will develop new AI methods to integrate
single-cell transcriptomics, epigenomics, and spatial omics data sets across
cancer types using large-scale public and in-house single-cell data. Next,
we will use these AI models to identify vulnerabilities in gene regulatory
networks and unravel the logic of genomic enhancers, with applications to
cancer genome variation and gene therapy. Finally, we will validate our
predictions, and optimize our AI models, using tumor organoid models and
high-throughput screens, focusing on melanoma and hepatic cancer.