Predicting clonal mutations from tumour spatial transcriptomics data using a graph convolutional neural network

Spatial transcriptomics (ST) has revolutionized our understanding of tumour biology but inherently lacks information on the upstream somatic driver mutations. We developed a spatially-aware graph convolutional neural network approach (SPOT-Mut) that infers mutated (sub)clones directly from ST data. SPOT-Mut was trained on virtual ST slides with clones simulated from a large collection of existing RNA and matched DNA sequencing data. SPOT-Mut was first evaluated for predicting mutations in TP53, the most frequently mutated gene in human cancer. Precision and recall values exceeded 95% in most analysed cancer types and the model was validated on ST and matched DNA sequencing data from serial slices obtained from cutaneous SCC samples. SPOT-Mut was then trained on 62 frequently mutated driver genes, resulting in highly performing somatic mutation prediction models for 40 driver genes, including KRAS, KEAP1, CTNNB1 and APC. With the increasing availability of ST data and upcoming atlases, SPOT-Mut potentiates the localization of (sub)clonal alterations in TP53 and other common driver mutations.