LingoCell interprets gene perturbations through natural language and tissue niches

LingoCell, a single-cell and spatial foundation model, reveals synergistic disease responses in lung adenocarcinoma to jointly weakening alveolar differentiation programs and strengthening malignant programs.

• Why it matters: Single-cell foundation models often fail to translate expression-magnitude changes into biological language, to transfer reliably to sparse spatial data, and to explain how cells relate to tissue niches.
• What they did: The model aligns cell representations with language while keeping gene expression editable, enabling annotation, cross-batch integration, description generation, and disease-language readouts for single-gene and combinatorial in silico perturbations.
• The result: In lung adenocarcinoma and colorectal cancer, it uncovers disease-linked program interactions and niche-conditioned program differences, supporting prioritization of target combinations and stromal intervention programs.

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