AI Framework Scores Gene Importance Across 400 Studies to Reveal Hidden Disease Links
A new AI tool called SIGnature decodes gene importance in single cells, linking a severe COVID-19 signature to Kawasaki disease.
Summary
Scientists at Genentech developed SIGnature, a computational framework that uses AI-powered single-cell RNA sequencing models to rank how important each gene is within a given cell type. Unlike simply measuring how much a gene is expressed, SIGnature uses attribution scores that cut through technical noise and highlight regulatory genes. Researchers applied it to the MS1 monocyte signature — a poorly understood gene program active in severe COVID-19 and sepsis — and searched across 400 studies. They discovered the same signature is activated in Kawasaki disease, a rare inflammatory condition primarily affecting children. Lab experiments confirmed that serum from Kawasaki disease patients can trigger the MS1 response. This suggests shared inflammatory mechanisms across seemingly unrelated diseases and opens doors to repurposing treatments.
Detailed Summary
Understanding which genes truly matter in a disease context has been a persistent challenge in genomics. Raw expression levels can be noisy and misleading, obscuring which genes are functionally driving a cellular state. A new AI-based framework called SIGnature aims to solve this problem at scale.
Researchers from Genentech and Boston Children's Hospital developed SIGnature to extract attribution scores from single-cell RNA sequencing (scRNA-seq) foundation models — large AI systems trained on vast amounts of gene expression data. These attribution scores reveal how much weight the AI places on each gene when characterizing a cell state, effectively ranking gene importance in a biologically meaningful way.
The team applied SIGnature to interrogate the MS1 monocyte signature, a gene program associated with severe COVID-19 and sepsis but poorly understood mechanistically. Searching across 400 published scRNA-seq studies, the framework identified the MS1 signature in multiple hyperinflammatory conditions. Critically, it surfaced Kawasaki disease — a pediatric inflammatory syndrome — as sharing this molecular program.
Experimental validation confirmed the connection: serum collected from Kawasaki disease patients was shown to induce the MS1 phenotype in monocytes in the lab. This cross-disease convergence suggests overlapping immune dysregulation pathways that could be therapeutically targeted across conditions.
The implications are broad. SIGnature could accelerate drug repurposing by identifying diseases that share underlying gene programs, even when their clinical presentations appear unrelated. It also offers a more principled way to compare datasets across research groups, addressing a longstanding reproducibility challenge in single-cell genomics. Caveats include the study's reliance on computational inference, and the Kawasaki disease validation, while compelling, was conducted with serum stimulation rather than in vivo disease models. Broader clinical translation will require prospective studies.
Key Findings
- SIGnature attribution scores outperform raw expression levels for ranking functionally important genes in single cells.
- Searching 400 scRNA-seq studies linked the MS1 monocyte signature to Kawasaki disease and other hyperinflammatory conditions.
- Kawasaki disease serum experimentally confirmed to induce the MS1 monocyte phenotype in vitro.
- The framework enables rapid cross-dataset gene set searches across large single-cell atlases.
- Shared molecular signatures across COVID-19, sepsis, and Kawasaki disease suggest common therapeutic targets.
Methodology
The study developed SIGnature, a computational package applying attribution methods to pre-trained scRNA-seq foundation models to score gene importance. Researchers performed a large-scale search of 400 published single-cell studies using the MS1 monocyte gene signature as a query. Experimental validation involved stimulating cells with serum from Kawasaki disease patients to assess MS1 phenotype induction.
Study Limitations
This summary is based on the abstract only, as the full paper is not open access. The experimental validation of the Kawasaki disease link relied on in vitro serum stimulation rather than in vivo disease models, limiting causal conclusions. Several authors have competing interests as Genentech or Roche employees, which warrants consideration when evaluating conclusions.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.