Senior ML Research Engineer, SandboxAQ

• Academic Foundation: Bachelor's degree in a scientific or quantitative field (Computer Science, Physics, Mathematics, Biology, Chemistry, or related); an advanced degree (MS or PhD) is preferred
• Applied ML in Science: Demonstrated experience building and maintaining machine learning models in a scientific discipline in an industry setting, including taking models from prototype through validation and maintenance; experience in a life sciences setting is preferred
• Large-Scale Data Management: Required experience managing large-scale datasets and managing the training of models over them, including data ingestion, cleaning, versioning, and pipeline maintenance at scale
• Software Engineering: Strong Python programming skills and experience with modern ML frameworks (e.g. PyTorch, JAX) and experiment tracking/data versioning tools (e.g. Weights & Biases)
• Scientific Rigor: Ability to design sound evaluation methodology (e.g. train/test splitting strategies, held-out generalization tests) and to critically interpret model performance against meaningful baselines

Desirable

• The AI Sim R&D team builds leading-edge ML and physics-based models ("LQMs") to advance drug discovery
• Within this team, AQCell is our virtual cell platform: it takes a cell representation (e.g. basal gene expression) and a perturbation descriptor (e.g. SMILES, dose, and time), predicts the resulting transcriptomic response, and maps that response through pathway activity to functional endpoints such as cell viability, IC50, and toxicity dose-response — helping drug discovery scientists understand the biological repercussions of a compound across the cell, not just whether it binds its target
• As a Machine Learning Engineer on AQCell, you will build and maintain the models and data infrastructure that power this pipeline
• You will work across large, heterogeneous transcriptomic and functional-endpoint datasets (e.g. LINCS L1000, GDSC, Tahoe-100M, and DILImap), train and evaluate expression-perturbation and cell-viability prediction models over them, and help harden our evaluation pipeline and baselines so we can trust and improve model performance over time
• This role sits at the intersection of machine learning and biology, and offers the opportunity to shape how virtual cell models are trained, validated, and scaled toward real drug discovery decisions
• Model Development: Build, train, and maintain machine learning models for expression-perturbation prediction (e.g. transcriptomic response to a drug perturbation) and downstream functional-endpoint prediction (e.g. cell viability, IC50, toxicity dose-response)
• Large-Scale Dataset Management: Acquire, harmonize, and manage large-scale biological datasets (e.g. LINCS L1000, GDSC, Tahoe-100M, DILImap) — including schema harmonization, normalization, and de-duplication across cell, drug, and assay identifiers — and manage model training pipelines over these pooled datasets
• Evaluation & Baselines: Contribute to automating and hardening the end-to-end evaluation pipeline, including implementing robust statistical baselines (e.g. cell- and drug-conditioned mean baselines) to rigorously benchmark model performance
• Research Translation: Translate ideas from the scientific literature (e.g. transformer-based perturbation models, knowledge graph and GNN-based embeddings) into working, well-tested code integrated into our modeling framework
• Cross-Functional Collaboration: Partner with computational biologists, software engineers, and product stakeholders to ensure models are grounded in sound biology and are usable in real drug discovery workflows
• Communication: Clearly document methods, assumptions, and results, and communicate findings to both technical and non-technical stakeholders