AI Biologist - SpatialBench-Long

AI Biologist - SpatialBench-Long

--------------------------------------

As molecular data generation and frontier model intelligence grows, new approaches to data analysis are needed across the biotech industry. Latch is building intelligent, high-performance agents for biological data analysis, empowering over 5,000 scientists across 150+ R&D labs to handle data from instrument-to-insights.

We're seeking a Bioinformatics Engineer to join our Spatial team, working at the frontier of what artificial intelligence can achieve in biology.

About the Role

You design tasks that test whether AI agents can work through real spatial transcriptomics projects. It's not about whether they can run a tool, it's whether they understand how decisions cascade. You pick how to align tissue images, which affects how you segment cell boundaries, which determines which unmixing method works, which shapes everything downstream. We're testing if they get that ripple effect. You'd be pulling from real published papers, building multi-stage tasks, and grading whether agents understand the *why* behind each choice, not just the mechanics.

Requirements

• Systems thinking across spatial workflows: You understand how image preprocessing (tissue alignment, background removal) affects segmentation quality; why nucleus-level segmentation demands different deconvolution (RCTD, Tangram, SPOTlight, STRIDE) than spot-level data; when morphological features matter for downstream inference
• Domain-specific judgment: You know the trade-offs between platforms (Visium's spots vs. MERFISH's near-cellular resolution vs. Xenium's high-plex imaging); when to deconvolve spots vs. when to segment first; how cell density and tissue morphology constrain method choice
• Tool flexibility: You learn spatial tools (Baysor, STdeconvolve, giotto, Squidpy) and image-processing pipelines as needed
• Task design: You build multi-checkpoint benchmar