Bioinformatics Engineer Variant USA

Human genetics is one of the most method-dense fields in biology. At every stage — from aligning sequencing reads to interpreting a fine-mapped locus to making a causal inference claim — a scientist is sitting between noisy population data and a judgment call that shapes what gets followed up and what gets dropped. VariantBench is our attempt to find out whether AI can make those calls.

The eval format is strict: data in, structured output out, graded against known ground truth. Every problem has a verifiable correct answer, no LLM judging LLM. The core skill we’re hiring for is the ability to design excellent questions. That means questions that test biological reasoning, not implementation details. Whether you used SuSiE or FINEMAP doesn’t matter. Whether you understand what the credible set tells you about the causal variant does.

We’re starting with population-scale analysis: GWAS, fine-mapping, colocalization, Mendelian randomization, where published summary statistics are rich enough to build rigorous, falsifiable evals. The more a candidate has ranged across the full stack: variant calling through QC through functional annotation through clinical interpretation, the more valuable they are to where this bench is going.

We need someone who has personally taken genetic association data and influenced what a result actually means; for a gene, a pathway, a patient. Not just someone who ran the pipeline. Someone who understood what came out the other end.