An evidence-governed agent harness for spatial transcriptomics prioritizes ICAM-1 as a target in lung fibrosis

DeepMAST identified ICAM1 as a candidate therapeutic target in idiopathic pulmonary fibrosis (IPF) Visium spatial transcriptomics, supported by spatial analysis and curated external evidence.

• Why it matters: Spatial transcriptomics yields many candidate targets, but functional validation needs cross-domain expertise and verification across biological resources.
• What they did: DeepMAST is an evidence-governed multi-agent AI system that orchestrates spatial analysis using MCP-enabled tools and locally deployed LLM Skills with RAG and guardrails. It was evaluated on controlled benchmarks and applications to prostate cancer, breast cancer, and IPF Visium datasets.
• The result: In the IPF application, ICAM1 nomination was supported by human-tissue immunostaining and ICAM-1 blockade in a preclinical fibrosis model, and DeepMAST improved evidence retrieval and reduced unsupported interpretations versus baselines.