MorphoProt: RNA-Conditioned Integration of Multiscale Histology for Spatial Protein Inference

MorphoProt predicts spatial protein abundance from RNA and multiscale histology, reaching mean proteinwise Spearman correlations of 0.652 in human tonsil, 0.604 in mouse spleen, and 0.372 in human liver.

• Why it matters: Spatial protein maps are costly and incomplete, while RNA alone misses cellular states and tissue organization that morphology can provide for protein inference.
• What they did: The model conditions RNA-guided cross-attention on five histological fields of view and refines fused features using gated aggregation from physical and transcriptomic neighborhoods, then predicts proteins.
• The result: On held-out sections, MorphoProt outperforms matched RNA-only, graph-based, HIPI, and SPRINT baselines in all three tissues, and combining fields of view improves accuracy while separating spleen and liver compartments and recovering germinal centers.

AI summary of the abstract · may contain errors · not medical advice · read the paper.