AI Biologist - Structural Biology

AI Biologist
- Structural Biology

Latch builds rigorous scientific benchmarks (benchmarks.bio) for AI agents across biology. We work with frontier labs and pharma to measure whether agents can handle real biological workflows: the messy, multi-layer reasoning that matters in the field.

We're seeking an AI Biologist to join our Structural Biology team, pushing what AI can achieve in interpreting macromolecular structure.

About the Role. You will be the scientific authority on ground truth, testing whether agents can reason like an expert structural biologist. AI agents today read a deposited model as fact. They miss that a ligand sits in empty density, that the two chains in a crystal file are a packing contact rather than the biological dimer, that a confident Alpha Fold prediction shows the wrong conformational state, or that a stretch of chain is built out of register.

Separating what the data actually support from what the model asserts is the core of the job. Knowing where the evidence ends, and when a structure cannot support a claim, is part of the work.

The role is intentionally broad, spanning molecular systems, experimental and computational methods, and questions from structure determination to interaction analysis, molecular dynamics, and functional interpretation. You will work with software engineers and scientists to build benchmark tasks and datasets across viral and bacterial proteins, toxins, RNA elements, and host–pathogen complexes. Your judgment will set the standard for agent outputs across these areas, including:

• Define what expert reasoning looks like for a structural question, and write gold-standard answers with the rationale an experienced crystallographer, cryo-EM scientist or biophysicist would defend.

• Design realistic scenarios across the structural workflow: data collection and processing, map interpretation, model building, validation, assembly and state assignment, comparison, functional interpretation, protein–protein interaction analysis, molecular dynamics simulation, and meta dynamics.

• Curate datasets across modalities: X-ray structure factors and maps, cryo-EM maps and half-maps, atomic models, Alpha Fold predictions with confidence metrics, molecular dynamics and meta dynamics simulations and trajectories, NMR shifts and ensembles, SAXS curves, HDX-MS, cross-linking, and SHAPE/DMS reactivity.

• Build analysis pipelines in Python (gemmi, Biopython/biotite, numpy) that compute verifiable ground truth directly from the files the agent receives.

• Partner with software engineers to turn your judgment into new workflows.

• Test whether agents distinguish real structure from artifact or over-interpretation: unsupported ligands, register shifts, mirrored maps, crystal contacts read as interfaces, misdocked domains, and B-factors or occupancies treated as evidence.

• Adjudicate ambiguous cases: decide what the map or measurement supports, where reasonable experts would disagree, and which additional data, such as a higher-resolution class, a solution-state measurement or a mutant, would settle the call.

• Test whether agents choose the right representation and comparison for the question: the biological assembly versus the asymmetric unit, the right reference state, and which domain to superpose on.

• Interrogate agent reasoning step by step, and challenge conclusions that sound right but would not survive peer review or a validation report.

• Pinpoint where agents go wrong: why a step fails, what an expert would do instead, and what evidence would change the call.

• Score agent confidence, rewarding explicit uncertainty over overconfident calls on ambiguous structures.

• PhD in structural biology, biophysics, biochemistry, or a…