Position Summary
The Consortium for Brain Research and AI in Neurodegeneration (C-BRAIN) is an academic-industry consortium directed by Dr. Randall J. Bateman, Washington University School of Medicine in St. Louis. C-BRAIN brings together academic research institutions, pharmaceutical and biotech companies, technology partners, philanthropic organizations, and government agencies to develop AI-powered tools to accelerate discovery in Alzheimer's disease and neurodegenerative disorders. Washington University serves as the coordinating center for C-BRAIN. C-BRAIN is currently deploying initial versions of the tools and developing plans for benchmarking and evaluation of the tools in addition to platform integration and scaling. C-BRAIN is also working on development of foundational AI models for brain aging and neurodegeneration. See www.c-brain.org.
Job Description
Primary Duties & Responsibilities:
Foundation Model Research and Development
- Lead the design of foundation models for spatial multi-omic brain data: encoders that handle different measurement panels across datasets, spatial and cross-scale context from subcellular to regional levels, and fusion of transcriptomic, genomic, proteomic, and imaging longitudinal measurements.
- Define and implement self-supervised training objectives (for example, masked measurement prediction, spatial-neighbor prediction, and cross-scale prediction), and determine which data modalities add the most information.
- Extend models to brain imaging and longitudinal clinical and biomarker data, and design how independently trained expert models combine into an assembly of experts.
- Prepare training-ready datasets with C-BRAIN data engineers: quality control, batch-effect assessment, harmonization across sites and platforms, and train, validation, and held-out splits that prevent leakage.
- Track the research field and adapt published methods and open models where they fit.
Scaling Studies and Training Infrastructure
- Design and run controlled scaling studies across data volume, compute, and model size; fit scaling curves; and recommend when larger models are justified.
- Train models from tens of millions toward billions of parameters on multi-node GPU clusters using distributed training (for example, PyTorch FSDP, DeepSpeed, or Megatron-Core) and schedulers such as Slurm.
- Build efficient data pipelines for very large imaging and spatial datasets, and measure GPU utilization, throughput, and cost so results reflect the model rather than infrastructure bottlenecks.
- Maintain reproducible training: versioned data and code, experiment tracking, checkpoints, and documented configurations.
Evaluation and Biological Validation
- Build benchmarks and evaluation suites with domain scientists: embedding geometry analyses, recovery of known biological relationships, and downstream tasks such as cell-type identification, disease staging, and brain-age prediction.
- Compare models against published baselines and simpler methods, and report results with uncertainty and clear limitations.
- Probe learned representations for new biological structure and work with scientists to test the resulting hypotheses.
Privacy-Preserving and Expert-Partitioned Training
- Train on contributed consortium data only through the privacy-preserving methods consortium data agreements require: per-contributor expert partitions, federated or confidential computation, and privacy protections applied to model parameters before they leave a protected environment.
- Evaluate privacy risk in trained models (for example, membership inference and data reconstruction) and apply mitigations such as differential privacy where needed.
- Work with C-BRAIN engineers and the partner-managed data and model-training platform on secure training environments, access approvals, and model release.
Model-to-Production Integration
- Package trained models for serving: optimization (for example, distillation and quantization), model cards, versioning, and evaluation gates before release.
- Work with C-BRAIN software engineers to expose models as tools that AI agents and researchers can call, with provenance on every result.
- Contribute to production inference code, including Rust-based serving where it improves performance or safety.
Technical Leadership and Scientific Communication
- Serve as technical lead for C-BRAIN modeling work: guide data scientists, data engineers, and research collaborators through design reviews, code review, and mentoring.
- Contribute to progress reports, Steering Committee materials, and grant reporting; publish and present results in line with consortium review procedures.
- Communicate model capabilities and limits clearly to scientists, partners, and non-technical leaders.
Working Conditions:
Remote within the United States, per current Washington University policies. Occasional on-site presence required for team meetings and consortium events. Occasional domestic travel may be required for consortium events, partner meetings, or scientific conferences. Standard office environment. Primarily sedentary with standard computer use.
Required Qualifications
Education:
PhD Or Terminal Degree Or Combination Of Education And Experience May Substitute For Minimum Education.
Work Experience:
Relevant Experience (3 Years)
More About This Job
Required Qualifications
PhD in machine learning, computer science, computational biology, statistics, or a related field, or a master's degree plus substantial equivalent research or industry experience training large models. Three years of experience after the PhD (or 6 years after a master's degree) building machine learning models on large datasets. Demonstrated experience designing and training large self-supervised or foundation models (transformer, graph, or comparable architectures). Hands-on distributed training on multi-GPU clusters with PyTorch (FSDP, DeepSpeed, Megatron-Core, or comparable). Experience with high-dimensional biological data: spatial omics, single-cell omics, biomedical imaging, or comparable scientific data. Production-quality Python with Git, code review, and automated testing. A track record of publications at major machine learning or scientific venues, or substantial open-source model or software contributions. Experience leading or mentoring others on technical work.
Preferred Qualifications
Training models at or above one billion parameters, or published scaling-law studies. Spatial transcriptomics or proteomics platforms (for example, CosMx, Xenium, MERFISH, CODEX, or imaging mass cytometry) and spatial analysis tools (for example, Squidpy and the scverse ecosystem). Mixture-of-experts or modular model architectures, including combining independently trained experts. Federated learning (for example, NVIDIA FLARE or Flower), confidential computing, or differential privacy. Large language models: fine-tuning, multimodal models that connect biological data with language, and models used as tools by AI agents. Rust, particularly for model inference and integration into production systems (for example, candle, burn, or ONNX Runtime). GPU-accelerated data and training stacks (for example, NVIDIA BioNeMo, Megatron-Core, RAPIDS, or DALI) and HPC environments with Slurm and containers. Brain imaging (MRI, PET) or longitudinal clinical and biomarker data. Knowledge of neurodegeneration or brain aging research (a plus, not a requirement). Experience in research consortia or with data governed by data use agreements.
Salary Range
$129,800.00 - $228,400.00 / Annually