Machine Learning of Single Cell Transcriptomic Data From anti-PD-1 Responders and Non-responders Reveals Distinct Resistance Mechanisms in Skin Cancers and PDAC

Frontiers in Genetics

Ryan Liu³  Emmanuel Dollinger⁴  Qing Nie⁴ 

[1] Center for Complex Biological Systems, University of California, Irvine, Irvine, CA, United States;Department of Developmental and Cell Biology, University of California, Irvine, Irvine, CA, United States;Department of Mathematics, University of California, Irvine, Irvine, CA, United States;NSF-Simons Center for Multiscale Cell Fate Research, University of California, Irvine, Irvine, CA, United States;

关键词: immunotherapy;    machine learning of single cell sequencing;    therapeutic response prediction;    supervised learning;    deep learning;    single-cell transcriptomic sequencing;   

DOI  :  10.3389/fgene.2021.806457

来源: DOAJ

【 摘 要 】

Immune checkpoint therapies such as PD-1 blockade have vastly improved the treatment of numerous cancers, including basal cell carcinoma (BCC). However, patients afflicted with pancreatic ductal carcinoma (PDAC), one of the deadliest malignancies, overwhelmingly exhibit negative responses to checkpoint therapy. We sought to combine data analysis and machine learning to differentiate the putative mechanisms of BCC and PDAC non-response. We discover that increased MHC-I expression in malignant cells and suppression of MHC and PD-1/PD-L expression in CD8+ T cells is associated with nonresponse to treatment. Furthermore, we leverage machine learning to predict response to PD-1 blockade on a cellular level. We confirm divergent resistance mechanisms between BCC, PDAC, and melanoma and highlight the potential for rapid and affordable testing of gene expression in BCC patients to accurately predict response to checkpoint therapies. Our findings present an optimistic outlook for the use of quantitative cross-cancer analyses in characterizing immune responses and predicting immunotherapy outcomes.

【 授权许可】

Unknown