RADAgent preprint reports local AI automation of genomic analysis workflow

Researchers report that RADAgent automated a complete genomic analysis workflow using an open-weight language model, keeping genomic data and model execution within a local computing environment. Their preprint evaluates the system on 476 samples from 19 populations. The workflow analyzes RADseq data, which samples subsets of genomes to genotype multiple individuals at low cost. RADAgent divides the analysis into 16 independently executable and verifiable modules, with controls for tool execution, checks of files on disk, and failure recovery. The evaluation covers one real population-genomics dataset; the researchers say the code is publicly available.