Multimodal protein representations across human brain cell types

SEMPER aligns protein language model embeddings with protein-protein interaction networks across 442 adult human brain cell types, enabling sequence-informed protein representations per expressed gene.

• Why it matters: Brain cell types are underrepresented in single-cell atlases and benchmarks, limiting how well models capture both protein sequence signals and cellular network context in the CNS.
• What they did: The authors train SEMPER on interactome data from 442 cell types and test it by predicting cell type-specific protein complexes and CRISPR screen essential genes in dopaminergic neurons.
• The result: SEMPER predicts dopaminergic neuron survival genes and, in zero-shot retrieval, recovers clinically pursued drug targets for Parkinson’s disease, Alzheimer’s disease, multiple sclerosis, Huntington’s disease, and amyotrophic lateral sclerosis at higher rates than cell type-specific models.

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