Principal Scientist — Human Brain Molecular World Model

The role

You will lead the development of Bexorg's human brain world models: generative models that predict how the molecular state of human brain tissue evolves over time, both in disease and in response to therapeutic intervention. You will take ownership of model architecture, training, and validation while helping establish priorities for the experimental data needed to improve model performance. Working closely with our drug discovery and development teams, you will help translate computational predictions into meaningful scientific and therapeutic decisions.

Your work will focus on Bexorg's proprietary, dynamic human brain molecular omics datasets, modeling molecular trajectories in continuous time across brain regions, cell types, pathways, and genes. This is a hands-on scientific leadership role with broad ownership and opportunities for cross-disciplinary collaboration.

What you will do

• Lead the architecture, development, training, and validation of generative human brain world models that predict molecular changes over time and in response to interventions.
• Develop models that capture high-dimensional molecular trajectories across brain regions, cell types, pathways, and genes.
• Work with Bexorg's proprietary human brain molecular omics datasets to improve model accuracy, biological relevance, and predictive capabilities.
• Partner with experimental teams to prioritize studies and data generation strategies that strengthen model capabilities and support scientific objectives.
• Apply advanced machine learning methods, including generative modeling, probabilistic learning, and approaches for high-dimensional dynamical systems.
• Incorporate biological inductive biases, uncertainty quantification, and structured inference into model development.
• Translate model predictions through Neurolens into actionable objectives for compound design and prioritization before synthesis.
• Collaborate with drug discovery and development scientists to inform target selection, compound advancement, dosing strategies, and donor-specific experimental contexts.
• Evaluate model capabilities, limitations, and uncertainty to guide scientific decisions and identify opportunities for additional experimentation.
• Develop and maintain reproducible machine learning pipelines with robust data and model lineage.
• Communicate complex computational findings clearly to multidisciplinary teams, supporting practical scientific and drug development decisions.
• Contribute to a collaborative research environment by working across disciplines, sharing expertise, and supporting the continued advancement of Bexorg's platform.

What you need

• A PhD in machine learning, computational biology, statistics, physics, or a related quantitative field.
• Postdoctoral or industry research experience.
• Demonstrated experience applying deep learning to solve complex problems in physics or biology.
• Hands-on experience with generative modeling approaches, including diffusion or flow-based methods, and modern probabilistic learning.
• Experience developing models that capture high-dimensional dynamical systems and incorporate symmetries or inductive biases.
• Strong understanding of structured, high-dimensional, low-sample inference, including hierarchical and partially pooled models, shrinkage methods, and uncertainty quantification.
• Proficiency in Python and experience developing machine learning models using PyTorch or JAX.
• Experience with distributed training and building reproducible machine learning pipelines with full data and model lineage.
• Hands-on experience running and troubleshooting machine learning workloads in cloud environments.
• A research publication record at venues such as NeurIPS, ICML, ICLR, or in journals such as Nature, Science, or Cell.
• Ability to communicate model capabilities, limitations, and scientific implications clearly to multidisciplinary teams.
• Willingness to collaborate across scientific disciplines, learn new domains, and contribute to broader research and platform development efforts.

Nice to have

• Experience with large-scale self-supervised learning or foundation models in scientific domains.
• Familiarity with modern causal inference and counterfactual reasoning, including their applications to biological or scientific problems.
• Experience with experimental design, active learning, value of information, or optimal experimental design.
• Biological fluency sufficient to translate model uncertainty into well-defined experimental questions.
• Familiarity with AWS compute, S3, and Cloud Development Kit (CDK).
• Experience with single-cell or spatial transcriptomics, including an understanding of the limitations of standard analysis pipelines.
• Knowledge of perturbation biology, including chemical or genetic screens, dose-response relationships, target engagement, or pharmacokinetics.
• Background or experience in neuroscience or neurodegenerative disease research.
• Experience with multimodal modeling across imaging and omics datasets.