bioRxivJClub
Predicting clonal mutations from tumour spatial transcriptomics data using a graph convolutional neural network
bioRxiv · · Preprint · Open access
Luijts, Hoogstoel + 10 more
Abstract ↗AI summary
The abstract is read at the publisher; the summary is JClub's.
SPOT-Mut predicts mutated subclones from spatial transcriptomics, achieving precision and recall exceeding 95% for TP53 across most analyzed cancer types using spatially aware graph convolution.
- Why it matters: Spatial transcriptomics lacks upstream somatic driver mutation information, creating a gap for localizing (sub)clonal alterations directly from ST without matched DNA sequencing.
- What they did: The model was trained on virtual ST slides with simulated clones from a large collection of existing RNA and matched DNA sequencing data, then evaluated on TP53 and validated on serial cutaneous SCC slices.
- The result: Training on 62 driver genes produced high-performing mutation prediction models for 40 genes, enabling localization of TP53 and other common driver mutations from ST data.
The findingWhy it mattersWhat they didThe result
- Open access