Staff Data Scientist, Imaging

Biohub•Redwood City, CA

4d•Hybrid

About The Position

Biohub is a large-scale initiative focused on accelerating scientific discovery by integrating frontier AI models, massive compute, and experimental capabilities. The AI research team leverages state-of-the-art AI to drive insights in biology, developing novel AI models, engineering robust systems, and creating practical tools for researchers. The approach is comprehensive, combining AI model development, engineering, biological data, computing infrastructure, and partnerships. Success relies on training biology-specific AI models, building efficient engineering systems, executing a data strategy, operating an AI compute platform, and creating accessible scientific tools. This role is within the Data team, responsible for the strategy, sourcing, and implementation of data for AI research and development. The goal is to enhance the speed, agility, and capability of biological AI research by connecting public data and Biohub's experimental platforms to AI systems. Biological data comes in various modalities (sequences, images, spatial coordinates, etc.), each with unique characteristics. Representing this data for learning is a key challenge. The role offers broad scope and high autonomy, influencing roadmap decisions and mentoring senior individual contributors. Success involves scaling data systems that are adaptive, interpretable, and scientifically grounded, accelerating progress toward biological frontier models and advancing human health. The ideal candidate will have a deep understanding of biological measurement, creative thinking about data representations and tokenization, and the ability to translate these into novel training architectures. Collaboration with experimental and computational scientists, data scientists, AI researchers, and data engineers is essential. This is an opportunity to invent methods for biological frontier models.

Requirements

PhD in computational biology, bioinformatics, or a quantitative biological field
Experience with tokenization strategies for non-text data (images, sequences, graphs, time series)
Track record of novel methodological contributions (publications, open-source tools, or production systems)
Familiarity with biological foundation models (ESM, scGPT, or similar)
Deep understanding of imaging data, their underlying data characteristics, and how to transform raw data into ai-ready datasets.
Experience designing data representations or feature engineering for machine learning, ideally in scientific or biological contexts
Familiarity with modern ML architectures (transformers, diffusion models, or similar) and how data representation choices affect learning
Strong computational skills (Python, scientific computing libraries); comfort working with large-scale datasets
Creative, first-principles thinking about how to structure data for learning

Responsibilities

Design data representations and tokenization strategies for imaging data that enable novel model architectures
Coordinate Experimental, Data Science, Data Engineering and AI Research teams to translate biological structure into learnable representations—defining priorities and appropriate structures for metadata and data that information models can access and consume
Guide data acquisition priorities, define quality criteria, and assess external datasets from a representation perspective
Develop and validate approaches for combining heterogeneous data modalities into unified training frameworks, designing for robustness to noise, bias, and batch effects
Evaluate how representation choices impact model performance, identifying which biological signals are captured or lost and iterating to improve