Tumor phenotype changes, such as trans-differentiation in lethal prostate cancers and hormone receptor
conversions in breast cancer, are increasingly frequent observations as resistance mechanisms to targeted
therapies. Therefore, characterizing the transcriptional regulation that drives treatment-induced tumor phenotype
changes during therapy in “real-time” has critical implications for studying mechanisms of resistance to therapies
and informing clinical treatment decisions. Surveillance of molecular changes in tumors is especially challenging
because the location and number of metastatic sites make it intractable to perform repeated biopsies. As a result,
it is difficult to characterize tumor evolution and cellular plasticity during therapy, exemplifying a major limitation
of current treatment strategies and precision medicine for patients with metastatic cancer. Circulating tumor DNA
(ctDNA) released from tumor cells into the blood is a non-invasive “liquid biopsy” solution for addressing
challenges in tissue accessibility. Current research and clinical efforts have focused on detecting genomic
alterations in ctDNA. However, studying the tumor phenotype from ctDNA remains challenging and is still a
nascent area of research.
The objective of this proposal is to develop an innovative computational method to profile and integrate genomic
alterations, chromatin accessibility, and transcriptional regulation directly from standard ctDNA sequencing data.
Recent advances and our preliminary studies now demonstrate the intriguing possibility to profile these “multi-
omic” patterns solely from computational analysis of standard ctDNA whole genome sequencing data. However,
there is still a lack of tools to predict transcriptional profiles from ctDNA. In Aim 1, we will develop a generalized
framework to predict transcriptional regulation from ctDNA. We will optimize ctDNA data normalization and
develop an unsupervised probabilistic generative model for predicting chromatin accessibility and transcriptional
regulation in ctDNA. To evaluate the method, we will perform benchmarking using plasma ctDNA from patient-
derived xenograft models. In Aim 2, we will test the hypothesis that the multi-omic signatures profiled from ctDNA
will provide a non-invasive approach to classify tumor subtypes and to survey molecular phenotype changes
during therapy. We will develop classifiers for predicting tumor subtypes and phenotype changes in adult and
pediatric cancers. To test the utility for characterizing multi-omic signature and predicting treatment-induced
phenotype changes, we will analyze serial ctDNA samples from patients receiving targeted therapies.
The method will be implemented as an open-source R package, and a workflow that can be deployed on local
and cloud environments, facilitating its adoption in the cancer research community. This proposal addresses the
urgent unmet clinical need for better analytical approaches to study cancer treatment resistance in “real-time”
and to advance cancer precision medicine.