GeneVariate is an open-source desktop application that uses a locally-run large language model (LLM) to explore how biological and experimental variables shape gene expression distributions across public microarray data. It works with any GPL platform in the Gene Expression Omnibus (GEO) across all organisms. The integrated LLM automatically extracts structured labels (tissues, conditions, treatment, treatment time, and age) from unstructured GEO metadata, enabling researchers to build targeted sample cohorts without manual annotation.
Its central function is identifying which variables drive samples into specific regions of a gene’s expression distribution. Researchers select any range such as a high-expression tail, bimodal peak, or low-expression sector, and enrichment testing reveals which biological groups are over-represented. Density and frequency visualizations show where label groups concentrate relative to the platform background. These patterns can be compared across genes, species, and platforms to uncover shared regulatory profiles or quantify technical and biological distances between experiments.
