Scoring gene importance by interpreting single-cell foundation models.

Area of Science:

• Computational Biology
• Genomics
• Immunology

Background:

• Assessing gene functional importance in cells is difficult, as expression levels are unreliable.
• Single-cell RNA sequencing (scRNA-seq) generates complex data, posing challenges for noise reduction and cross-dataset comparisons.

Purpose of the Study:

• Introduce SIGnature, a novel framework for scoring gene importance using scRNA-seq attributions.
• Develop a tool for gene set searches across large scRNA-seq datasets.
• Uncover shared biological mechanisms across different diseases.

Main Methods:

• Utilized foundation models for scRNA-seq data to derive gene attribution scores.
• Developed the SIGnature R package for generating and querying gene attributions.
• Applied SIGnature to identify associations with the MS1 monocyte signature in 400 studies.

Main Results:

• Attribution scores effectively reduce technical noise and highlight regulatory genes.
• SIGnature facilitated cross-dataset comparisons, a key challenge in scRNA-seq analysis.
• The MS1 monocyte signature was associated with hyperinflammatory conditions, including severe COVID-19, sepsis, and Kawasaki disease.
• Experimental validation confirmed that serum from Kawasaki disease patients induces the MS1 phenotype.

Conclusions:

• SIGnature provides a powerful approach for large-scale gene signature scoring and cross-disease analysis.
• The framework can uncover shared molecular mechanisms underlying diverse conditions.
• Identified novel associations between the MS1 signature and hyperinflammatory diseases, with potential diagnostic and therapeutic implications.