In Data Analysis We Trust

Abstract

Workflow languages such as Nextflow and Snakemake are a great way to ensure reproducibility for the preprocessing steps in bioinformatics. But what about the downstream steps? With LLMs you can write analysis scripts much faster than ever, so how do you make sure the results are correct? This talk shares my experience using LLM-assisted coding for downstream genomics analysis, the failure modes I have hit, and how to keep results trustworthy when AI is in the loop.