Drugging tumor heterogeneity
Although genomics technologies have advanced, tools for interpreting the resulting data have not kept pace and thus the influence of tumor heterogeneity on cancer progression and drug response is not fully understood. Our lab is focused on developing generalizable computational approaches that can be applied to understand tumor heterogeneity, then nominating new tumor vulnerabilities by the intersection of these data with data from large preclinical screening efforts. We use these insights to design comprehensive studies of the most promising novel therapeutics and drug combinations, with a view to improving outcomes primarily in high-risk pediatric neuroblastoma.
In the last 5 years, vast strides have been made in our ability to study patient tumor heterogeneity using high-throughput genomics methods like single cell and spatial transcriptomics. In parallel with these developments, we have gained the ability to screen very large numbers of preclinical disease models (including cancer cell line panels) with pharmacological agents or CRISPR/Cas9 gene editing. However, our ability to jointly leverage these new classes of data is lacking. For example, while screening in cancer cell lines is highly scalable and has yielded clinically impactful findings across many diseases, the fidelity of these models is often unclear and the computational infrastructure to determine which (if any) components of primary tumors are recapitulated by various preclinical models has not been developed. The overall goal of our lab is to develop the computational infrastructure that allows the results from preclinical screening to be interpreted in the context of a detailed understanding of patient tumor biology, and in understanding this relationship to nominate new opportunities for drug repurposing, the development of new drug targets and new drug combinations. The primary computational tools we use for this purpose are that of unsupervised and supervised machine learning. While the computational work is highly generalizable, we specifically use these unique computational skills to push forward our own clinical translational and wet-lab mechanistic work focused on high-risk neuroblastoma.