Alpha Spatial: Language-guided prediction of single-cell spatial gene expression from histology
Routine histology preserves tissue architecture but leaves its molecular organization largely unmeasured. Recovering this information requires learning across heterogeneous gene panels and constraining a biologically ill-posed morphology-to-expression mapping. Here we present Alpha Spatial (AS), a language-conditioned framework that predicts cell-level gene expression from histology while jointly…
Alpha Spatial is a language-guided framework that predicts cell-level gene expression from histology, while learning from diverse tissues and panels. By incorporating matched biological context, the framework enhances prediction accuracy, whereas incompatible context decreases it. The use of multi-tissue learning improves prediction in breast tissue, and AS outperforms existing methods while maintaining molecular-state organization.
In breast cancer, AS successfully recovers stromal programs associated with POSTN and reveals HER2-associated molecular and spatial differences. Through patient-level explorations, AS identifies a CXCL14-associated spatial state linked to non-response to trastuzumab-containing treatment and larger residual invasive tumors. This study demonstrates a novel approach to leveraging heterogeneous molecular observations from archived histology to understand tissue organization and treatment-associated microenvironments.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
