Post by Marcellus Augustine

Very excited to share our latest research, published in Nature Machine Intelligence. We describe our approach to ‘Immunotherapy drug target identification using machine learning and patient-derived tumour explant validation’. https://lnkd.in/eqARTXNd Despite the revolutionary impact of immune checkpoint inhibitors (CPI) on cancer treatment, only a minority of patients experience durable clinical responses. To address this urgent need for novel therapies that can rescue patients for whom existing CPIs are ineffective, we developed MIDAS (Mining Immunotherapy Drug tArgetS). MIDAS is a multimodal graph neural network (GNN) system dedicated to cancer immunotherapy target discovery. It harnesses highly complex, multi-dimensional data to profile tumour-immune dynamics across complementary lenses: multi-omic patient molecular profiles from CPI-treated cohorts, single-cell transcriptomics atlases of tumour-infiltrating immune cells, functional evidence from gene perturbations in CRISPR tumour-T cell co-cultures, HLA-peptidomics data, and population-scale genetic associations. MIDAS robustly identifies held-out known immunotherapy targets, generalises to time-sliced data, and differentiates approved targets from those in clinical trials. Explainability analyses reveal that MIDAS enriches for core tumour immunology pathways, draws on transcriptional dynamics, gene interactions, autoimmune function, Tregs, and functional dendritic cells. MIDAS highlighted the Oncostatin M-Oncostatin M receptor (OSM-OSMR) axis, previously explored in autoimmunity, as a candidate cancer immunotherapy target. Functionally perturbing OSM-OSMR in TRACERx patient-derived melanoma explants (PDEs) remodelled the tumour microenvironment, with a significant decrease in dysfunctional T cells and CCL-4 levels. Bottom line: Our findings show that GNNs can integrate multimodal, multi-omics datasets to accurately model tumour-immune dynamics and identify data-driven candidates that show promise in advanced preclinical patient-derived platforms. Leveraging tools such as MIDAS could help streamline the development of next-generation cancer immunotherapies. A massive thank you to my amazing co-authors: Nuno Rocha Nene, Hongchang Fu, Christopher Pinder, Lorena Ligammari, Alex Simpson, Irene Sanz, Krupa Thakkar, Dawen Qian, Evelyn Fitzsimons, Ben Simpson, Roberto Vendramin, Andrea Castro, Heather Niederer, Samra Turajlić, Sergio Quezada, Nicholas McGranahan, Chris Watkins, Charles Swanton, and Kevin Litchfield. Also, thanks to the UCL Cancer Institute, The Francis Crick Institute, Cancer Research Horizons, Cancer Research UK City of London Centre (CoLC), Cancer Research UK (CRUK) Lung Centre, Medical Research Council, CRUK CoLC Clinical Academic Training Programme, UCL MBPhD Programme, and the CRUK Therapeutic Catalyst Award.