Director of Molecular AI & Federated Learning

Eli Lilly and Company; Indianapolis (IN); On-site; USD 177,000 - 282,000; Full time

Job summary

The Director, Molecular AI & Federated Learning at Lilly TuneLab leads the technical vision for privacy-preserving federated learning and generative small-molecule design, combining medicinal chemistry, ADMET prediction, and advanced ML to accelerate lead optimization across a federated network.

The role requires PhD-level expertise, 5+ years in ML for drug discovery, and ability to mentor scientists, set strategy, and collaborate with internal teams and external biotech partners across

Qualifications

• PhD in a relevant computational field is required.
• 5+ years post-PhD in ML for drug discovery or equivalent leadership.
• Strong publication and collaboration skills.

Responsibilities

• Set the technical direction for federated learning and molecular AI across TuneLab—defining a research agenda that unifies privacy-preserving foundation models, multi-task learning, and generative small-molecule design, and aligning it with platform and portfolio priorities.
• Serve as a principal technical authority and mentor for data scientists and engineers—guiding experimental design, reviewing methods and code, and raising the scientific bar across the team, while influencing technical decisions across disciplines internally and with external partners.
• Architect novel deep learning architectures (e.g., Transformer and graph neural network-based) for large-scale federated pre-training on unlabeled or partially labeled data distributed across multiple partner sources.
• Advance state-of-the-art semi-supervised and self-supervised methods (e.g., contrastive learning, masked auto-encoding) tailored to the constraints of federated learning, such as communication bottlenecks and data heterogeneity.
• Develop robust, communication-efficient aggregation strategies (e.g., FedAvg, FedProx, SCAFFOLD) that remain stable for large, complex models and handle non-IID data across clients.
• Profile and optimize the computational performance—memory, latency, and communication cost—of federated training and inference for scale, and build high-fidelity simulation environments to test, debug, and benchmark federated strategies before real-world deployment.
• Architect multi-task learning models that leverage shared representations across related endpoints to improve predictive performance and data efficiency in a federated ecosystem, where each client may hold data for only a subset of tasks.
• Design algorithms that address extreme task and feature heterogeneity across clients—personalized models, meta-learning, and gradient-aggregation methods robust to non-IID data—and apply regularization that prevents negative transfer while encouraging positive knowledge sharing.
• Create efficient protocols for fine-tuning and adapting pre-trained federated models to specific downstream tasks, and establish rigorous validation frameworks with appropriate per-task metrics and fairness assessment across clients and tasks.
• Build multi-task models for small-molecule properties—including ADMET endpoints, solubility, permeability, metabolic stability, and off-target liabilities—across diverse chemical representations (SMILES, graphs, 3D conformations).
• Design and deploy state-of-the-art generative models (VAEs, diffusion models, flow matching, autoregressive models) for de novo design, lead optimization, and scaffold hopping that respect synthetic accessibility and drug-likeness constraints.
• Develop integrated prediction-generation pipelines that optimize molecules simultaneously across multiple ADMET properties while maintaining target potency, using multi-objective optimization and Pareto-front exploration.
• Implement efficient exploration of synthetically accessible chemical space—reaction-aware generation, retrosynthetic-planning integration, and fragment-based design—collaborating with synthetic chemists to ensure generated molecules are practically synthesizable.
• Learn and exploit structure-activity relationships from sparse, noisy federated bioactivity data—including matched molecular pair analysis and activity-cliff prediction—and develop self- and semi-supervised molecular representations that generalize to novel chemical series.
• Apply explainability (XAI) techniques to complex multi-task and molecular models to understand predictions and uncover relationships between endpoints, generating novel scientific insight while respecting IP and competitive boundaries across federated partners.
• Establish rigorous benchmarks using public and proprietary data; author publications and deliver presentations; uphold reproducible code and version control for data, code, and models.

Skills

Federated learning

Graph neural networks

Molecular AI

Multi-task learning

Generative chemistry

ADMET prediction

Transformers

Molecular design

Privacy-preserving ML

Mentorship

Education

PhD in Computer Science, Computational Chemistry, Cheminformatics, Machine Learning, Computational Biology, or a related computational field

Tools

RDKit

DeepChem

PyTorch

Job description

The Director, Molecular AI & Federated Learning at Lilly TuneLab leads the technical vision for privacy-preserving federated learning and generative small-molecule design, combining medicinal chemistry, ADMET prediction, and advanced ML to accelerate lead optimization across a federated network.

The role requires PhD-level expertise, 5+ years in ML for drug discovery, and ability to mentor scientists, set strategy, and collaborate with internal teams and external biotech partners across