Vibe analyzing my genome

A LessWrong user conducted a sophisticated personal genome analysis using AI assistants Claude and ChatGPT, working with their 43x whole-genome sequence data from Nucleus Genomics. The project involved multiple technical steps including pharmacogenomic star-allele calling via PharmCAT and Cyrius, HLA typing via OptiType, ClinVar annotation of 76 genes, and filtering 5 million variants down to 27 high-impact candidates. The analysis revealed the user has an extremely rare multi-CYP drug metabolism profile occurring in approximately 1 in 23,000 people, which explains their history of extreme multi-drug sensitivity.

The AI-assisted workflow included creating drug contingency tables covering 75+ medications with pathway-specific safety ratings and cross-referencing 23 specific medications against the user's genotypes. While the user acknowledges limited genetics knowledge, they employed rigorous validation including critical adversarial review of findings against population frequencies and ClinVar review status. This case demonstrates how LLMs can enable complex bioinformatics analysis for personalized medicine applications, though the author cautions about the risk of misinterpretation without careful validation.