DR-428 Transcriptomic and Genome-wide machine learning model for predicting immunotherapy benefit in metastatic solid tumors

Immunotherapy has radically changed the treatment landscape across solid tumors. Regardless, biomarkers able to predict immunotherapy benefit are limited, and mostly based on single pathological biomarkers scarcely capable of predicting consistently which patients might derive the most benefit from using Immune Checkpoint Inhibitors (ICI). Despite machine-learning models (MLM) have been built to predict ICI benefit, they have been largely based on retrospective dataset of limited sample size, with limited application in external validation cohorts and across different primary tumors. Moreover, they have been largely based on single-modality models, with preliminary data suggesting multi-modality features could better capture immunotherapy sensitivity. Accordingly, the present project aims to build a multi-modal MLM based on clinicopathological, genomic (assessed with either Whole-genome sequencing [WES] or Whole-genome sequencing [WGS]) and transcriptomics (RNASeq), which will be built and trained on the cohort of solid tumors treated with immunotherapy in the Hartwig dataset (n= ~450). Following, an external validation cohort of triple negative breast cancer treated with pembrolizumab plus cyclophosphamide within a phase II clinical trial (NCT03971045) (n= ~50) will be used as an external, independent validation cohort.

Zhan Yinxiu, Istituto Europeo di Oncologia, Italy