About this role
Bridging the gap between automated data generation and experimental biology, this role addresses the challenge of building reliable domain models for perturbation biology, genetics, and high-dimensional readouts. The Life Science AI team at Lila Sciences creates machine learning systems for automated reasoning, where models directly inform what biological experiments should be run next. Working closely with experimental scientists, the incoming specialist will design interpretable structures grounded in biological machinery rather than standard cell-type embeddings.
Candidates joining this team will focus on generalization under structured sparsity, creating predictions for unmeasured conditions while prioritizing calibrated uncertainties and honest error bars over simple embeddings. Ideal practitioners bring a rigorous academic background, deep familiarity with high-dimensional biological data, and a track record of steering ambiguous research projects from formulation to execution. Someone who thrives on connecting machine learning methods directly to biological mechanisms and therapeutic opportunities will find a strong fit here.
What you will do
- Build domain models for perturbation, genetic, and multimodal experimental data while translating biological questions into rigorous machine learning formulations.
- Develop models with interpretable structure and parameters that feature calibrated posteriors and honest error bars for unmeasured conditions.
- Pre-register simple entity-mean, additive, and linear baselines before fitting any model to ensure transparent performance metrics.
- Partner with experimental scientists to guide data generation, model validation, and experiment-selection methods.
- Design benchmarks connecting model performance to therapeutic consequences and apply outputs to prioritize targets and mechanisms.
- Support the integration of domain models into agentic workflows by deploying tools and helping engineer agent harnesses.
What you need
- Hold a PhD in machine learning, statistics, computational biology, computer science, bioengineering, physics, or a related quantitative field with a strong publication record.
- Demonstrate hands-on experience developing models for high-dimensional biological data and connecting methods to experimental design.
- Show a proven track record of generalization under structured sparsity, predicting into conditions not directly observed rather than interpolating densely sampled corpora.
- Bring comfort with calibration, proper scoring rules, and evaluation design for high-stakes decision-making environments.
- Possess strong programming skills, reliable machine learning research workflows, and clear cross-functional communication abilities.
Nice to have
- Mechanistic and probabilistic modeling including Bayesian hierarchical models or neural differential equations
- Experience with active learning, Bayesian optimal experimental design, or closed-loop experimentation
- Experience deploying research models into scientific decision-making workflows as callable tools
Benefits
- Comprehensive medical, dental, and vision coverage
- Employer-paid life and disability insurance
- Flexible time off with generous company-wide holidays and paid parental leave
- Educational assistance program and commuter benefits including bike share memberships
- Company-subsidized lunch program
Skills and tech stack
Questions candidates ask
What is the expected compensation range for this position?
The expected base salary range is $268,000 to $358,000 USD, with final offers reflecting individual background, expertise, and expected impact.
What kind of models will I be building?
You will build domain models for perturbation, genetic, and multimodal experimental data, translating biological questions into rigorous machine learning problem formulations with interpretable structures.
How does the team evaluate model performance?
Baselines are pre-registered before models are fit, and the team values calibrated posteriors, honest error bars, and structured generalization over simple embedding accuracy.