SAGE-FM Demonstrates Spatially Coherent Gene Embeddings

A Jan. 21, 2026 arXiv preprint introduces SAGE-FM, a lightweight spatial transcriptomics foundation model based on graph convolutional networks. Trained on 416 human Visium samples from 15 organs with a masked central spot objective, it recovers masked genes (91% significant correlations), enables 81% spot-annotation accuracy, and improves glioblastoma subtype prediction versus MOFA while capturing directional ligand–receptor regulatory effects.
Scoring Rationale
Strong methodological novelty and practical performance, offset by single-source preprint status and limited external validation.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems

