ML Researcher

About the role

Axiom is building a platform to replace animal and legacy toxicity testing with human-relevant predictive models. As a founding team member, you will create AI systems that determine if a molecule is toxic to humans and explain the underlying biological reasons.

Key facts

What you'll do

• Define the end-to-end research agenda for ML and agentic systems, from wet-lab data generation to customer-facing deployment.
• Build models that map the relationships between chemical structure, biological response, dosage, and human toxicity.
• Train large multimodal models using chemical structures, high-content cellular images, transcriptomics, proteomics, mass spectrometry, ADME, and clinical data.
• Develop foundation models and representation-learning systems for biological images and molecular data.
• Architect predictive models for human toxicity based on dose, Cmax, in vitro potency, and biological state.
• Create methods to align and interpret embeddings across diverse assays, timepoints, and biological modalities.
• Conduct rigorous error analysis to identify model failure modes and determine data requirements for improvement.
• Collaborate with chemists, computational biologists, and wet-lab scientists to design experiments that enhance model performance.
• Build mechanistic agents capable of reasoning over experimental data and literature to guide scientific decision-making.
• Manage the full research-to-product lifecycle, including prototyping, training, evaluation, and shipping.

Requirements

• Demonstrated history of exceptional machine learning work in industry, academia, open source, or independent projects.
• Strong engineering skills with the ability to write PyTorch code, debug training runs, and scale inference.
• Experience managing messy, sparse, and biased biological data.
• Ability to transition effectively between theoretical research and production-level systems.
• Willingness to learn the fundamentals of biology, chemistry, pharmacology, and toxicology.
• Focus on model evaluation, calibration, and real-world utility.

Nice to have

• Experience with multimodal ML across imaging, omics, and chemical data.
• Background in reinforcement learning applied to biology or chemistry.
• Expertise in uncertainty estimation and confidence calibration for scientific decisions.
• Knowledge of mechanistic interpretability for chemical or biological models.
• Experience building evaluation systems for real-world scientific problems rather than toy benchmarks.

Skills & tools

• Deep learning frameworks: PyTorch, JAX, TensorFlow.
• Scientific computing: Python, NumPy, Pandas, Polars, PyArrow, scikit-learn.
• Modeling techniques: Representation learning, contrastive learning, metric learning, self-supervised learning, graph neural networks, generative models.
• Specialized domains: Computer vision for biological imaging, high-content screening, cell painting.
• Infrastructure: Large-scale distributed training, GPU clusters, inference pipelines, cloud compute.
• Systems: LLMs, agentic workflows, retrieval-augmented generation, tool use.

Practical notes

Axiom currently serves 7 of the top 20 pharmaceutical companies and several innovative biotech firms. The role focuses on solving drug-induced liver injury as a primary use case, with plans to expand across all major human organ systems.