AI Biologist - Structure Biology

LatchBio•San Francisco, CA

4d•$120,000 - $180,000

About The Position

Latch builds scientific benchmarks for AI agents across biology and works with frontier labs and pharmaceutical companies to assess agents on real biological workflows. The AI Biologist will join the Structural Biology team and serve as a scientific authority on ground truth, evaluating whether agents reason appropriately about macromolecular structures and distinguish what experimental data support from what a deposited or predicted model asserts. The role spans molecular systems, experimental and computational methods, and work ranging from structure determination to interaction analysis, molecular dynamics, and functional interpretation. The position develops benchmark tasks and datasets across viral and bacterial proteins, toxins, RNA elements, and host–pathogen complexes, in collaboration with software engineers and scientists. This includes defining expert-level answers and rationales, designing realistic structural workflows, curating multimodal datasets, and building Python pipelines to calculate verifiable ground truth from source files. The AI Biologist will create challenging tests of artifact recognition, adjudicate ambiguous evidence, evaluate agents’ choice of representations and comparisons, and scrutinize their reasoning and confidence. The role is remote, hybrid, or onsite in San Francisco, with a preference for onsite work. It is a 1099 contract (or W-8BEN internationally) with no stated end date and a full-time commitment on the worker’s own schedule. Pay is guaranteed for weeks 1–4 during ramp-up and then based on certified tasks; most people in the position earn $120,000–$180,000 or more, with uncapped upside.

Requirements

PhD in structural biology, biophysics, biochemistry, or a related field.
Hands-on structural experience, including personally building and validating models and assessing map, model, and fit quality.
Deep expertise in at least one of X-ray crystallography, cryo-EM, NMR/solution methods, or computational structural biology, and working fluency across the others.
Ability to code for structural data and compute contacts, superpositions, symmetry mates, map values, and geometry from files without a GUI.
Daily use of Claude, Cursor, or similar AI agents, with the ability to identify reasoning that is technically correct but scientifically unsound.
Confidence to overrule an AI agent when evidence demands it and openness to update when shown wrong.
Ability to calibrate confidence and communicate uncertainty rather than defaulting to binary conclusions.

Nice To Haves

Structural virology or antibody–antigen structural work, including viral glycoproteins, epitope mapping, and structural interpretation of escape.
RNA structural biology or protein–RNA interaction analysis.

Responsibilities

Define expert reasoning for structural biology questions and write gold-standard answers with defensible scientific rationales.
Design realistic benchmark scenarios spanning data collection and processing, map interpretation, model building and validation, assembly and state assignment, comparison, functional interpretation, protein–protein interaction analysis, molecular dynamics, and metadynamics.
Curate datasets across structural and experimental modalities, including X-ray structure factors and maps, cryo-EM maps and half-maps, atomic models, AlphaFold predictions and confidence metrics, simulation trajectories, NMR data, SAXS, HDX-MS, cross-linking, and SHAPE/DMS reactivity.
Build Python analysis pipelines using tools such as gemmi, Biopython/biotite, and numpy to compute verifiable ground truth from the files agents receive.
Partner with software engineers to translate scientific judgment into new workflows.
Test whether agents distinguish genuine structural evidence from artifacts and over-interpretation, including unsupported ligands, register shifts, mirrored maps, crystal contacts mistaken for interfaces, misdocked domains, and misuse of B-factors or occupancies.
Adjudicate ambiguous structural cases, determine what evidence supports, identify reasonable expert disagreement, and specify what additional data could resolve a call.
Assess whether agents choose appropriate representations and comparisons for a question, such as biological assembly versus asymmetric unit, reference state, and domain for superposition.
Interrogate agent reasoning step by step and challenge conclusions that would not withstand peer review or validation.
Identify why agent reasoning fails, explain what an expert would do instead, and state what evidence could change the conclusion.
Score agent confidence and reward explicit uncertainty where structures are ambiguous.
Use AI agents in daily work while applying original expert judgment to their outputs.