Multi-omic machine learning predictor of breast cancer therapy response

Breast cancers are complex ecosystems of malignant cells and the tumour microenvironment. The composition of these tumour ecosystems and interactions within them contribute to responses to cytotoxic therapy. We collected clinical, digital pathology, genomic and transcriptomic profiles of pre-treatment biopsies of breast tumours from 168 patients treated with chemotherapy with or without HER2 (ERBB2) targeted therapy before surgery, and correlated pathologic outcomes (complete response or residual disease) with multi-omic features. We show that response to treatment is modulated by the pre-treated tumour ecosystem and that its multi-omics landscape can be integrated into predictive models using machine learning. The degree of residual disease following therapy is monotonically associated with pre-therapy features, including tumour mutational and copy-number landscapes, tumour proliferation, immune infiltration, and a genomic factor associated with response to HER2-targeted therapy. We validate predictive performance in an external cohort and demonstrate that integrating clinical, molecular and digital pathology features improves prediction of response. This approach could be used to develop predictors for cancers.

Stephen-John Sammut, Mireia Crispin-Ortuzar, Suet-Feng Chin, Elena Provenzano, Helen A Bardwell, Wenxin Ma, Veil Cope, Ali Dariush, Sarah-Jane Dawson, Jean E Abraham, Janet Dunn, Louise Hitel, Jeremy Thomas, David A Cameron, John M S Bartlet, Larry Hayward, Paul D Paroha, Florian Markowetz, Oscar M Rueda, Helena M Earl, Carlos Calderon