189P - An image-based deep learning prediction model for characterization of the drug tolerant persister cell state

Background

Targeted therapies improve the clinical outcome of cancer treatment. However, in solid tumors a cancer cell subpopulation survives during initial therapy and evolves into drug tolerant persister cells (DTPs) that maintain a residual disease reservoir. Residual disease contributes to lethal tumor progression; identifying and eliminating DTPs could benefit future treatment paradigms. We have shown that Hippo pathway effector Yes Associated Protein-1 (YAP1), a transcriptional co-regulator, is activated in oncogene-driven cancers in response to targeted therapy and maintains the DTP phenotype. In this study, we develop and validate a deep learning-based predictive model for the identification of active YAP1-mediated DTP states from hematoxylin & eosin (H&E) and immunohistochemistry (IHC) images of lung and skin cancer.

Methods

H&E and/or paired YAP1-stained IHC images from clinical models of lung and skin cancer were collected throughout targeted therapy and annotated for active YAP1-containing cells with semi-automation using high performance computing clusters. A modified U-Net algorithm was used for image segmentation, training, validation, and testing.

Results

2,267 images were annotated, producing over 80,000 patches containing YAP positive cells that comprised the training dataset. Subsequently, we constructed a customized deep-learning model to detect YAP positive DTP cell states from whole histopathological image slides. The model achieved 0.9091, 0.8949, and 0.902 accuracy in training, validation, and testing datasets for lung cancer, respectively. We applied our model to an external dataset of LUAD diagnostic H&E images (n=541) for YAP1 prediction and showed higher YAP1 scores correlate with poor overall survival.

Conclusions

Our model detects YAP1 activation-mediated DTPs throughout targeted therapy treatment. Following further clinical validation, model implementation into routine cancer care in the future could identify patient subpopulations with YAP1 activated tumors who would most benefit from receiving YAP1-targeted small molecule inhibitors.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

University of California, San Francisco.

Funding

National Science Foundation Graduate Research Fellowship Program, Helen Diller Comprehensive Cancer Center, University of California Discovery Fellowship.

Disclosure

T. Bivona: Financial Interests, Institutional, Research Funding: Novartis, Strategia, Kinnate, Revolution Medicines; Financial Interests, Personal and Institutional, Advisory Role: Array/Pfizer, Revolution Medicines, Springworks, Jazz Pharmaceuticals, Relay Therapeutics, Rain Therapeutics. All other authors have declared no conflicts of interest.