2027 Future Talent Program – Translational Sciences and Outsourcing – Co-op

Project Overview

The co-op will develop and benchmark structure-aware models that combine compound structures, physicochemical properties, and initial in vivo PK/TK data to predict exposure at untested dose levels based on public and internal data. The project will evaluate whether AI, ML, and mechanistic hybrid approaches improve prediction accuracy, uncertainty estimation, and detection of nonlinear exposure relative to standard empirical and mechanistic baselines.

Primary Responsibilities

• Curate and quality-check preclinical cross-dose PK/TK datasets with clear provenance, reusable data definitions, and appropriate holdout splits to construct reusable benchmark dataset.
• Conduct TK analysis with exposure endpoints such as AUC and Cmax for different compound, dose, and route of administration to identify trends, nonlinearities, and data limitations.
• Develop and benchmark AI-enabled and hybrid modeling approaches against relevant baselines such as linear scaling, power/Emax models, QSAR/ChemProp-based methods, and mechanistic approaches where applicable.
• Evaluate model performance, uncertainty estimation, and ability to detect nonlinear PK behavior such as saturation or sex differences.
• Create clear visualizations, written summaries, and technical presentations for cross-functional stakeholders.
• Share project outcomes through internal and/or external presentation and publication.

Required Education

Currently pursuing a Master's or Ph.D. degree in data science, computer science, applied mathematics, statistics, computational chemistry, chemical or biomedical engineering, pharmaceutical sciences, or a related quantitative discipline. Research experience in machine learning, statistical modeling, scientific computing, or computational modeling.

Required Skills

• Strong Python programming skills, including scientific computing with NumPy, pandas, and scikit-learn.
• Experience with deep learning frameworks such as PyTorch or JAX.
• Solid foundation in machine learning, statistics, model evaluation, and reproducible analysis workflows.
• Ability to work with messy tabular or time-series scientific data and translate analytical findings into clear plots and written summaries.
• Effective written and verbal communication skills and ability to collaborate in a cross-functional research environment.

Preferred Skills

• Exposure to graph neural networks, molecular representation learning, ChemProp, QSAR modeling, or related cheminformatics methods.
• Familiarity with ODEs, dynamical systems, differentiable simulation, torchdiffeq, Neural ODEs, or mechanistic modeling concepts.
• Background or strong interest in pharmacokinetics, toxicokinetics, pharmacometrics, PBPK, ADME, or preclinical safety assessment.
• Interest in agentic AI or LLM-enabled tooling for scientific workflow automation.
• Fluent in leveraging state-of-the-art AI tools for boosting productivity, such as Claude Code, ChatGPT, Gemini, DeepSeek, etc.

Learning Opportunities

The co-op will gain hands-on experience applying modern AI/ML methods to real pharmaceutical PK/TK data with direct impact on preclinical decision-making. The role will build technical depth in molecular representation learning, model benchmarking, uncertainty estimation, PK/TK data analysis, and scientific communication in a cross-functional pharmaceutical R&D setting.

Expected Deliverables

• A curated cross-dose preclinical PK/TK dataset with provenance, quality checks, and holdout splits suitable for reuse by follow-on efforts.
• A documented benchmark comparing AI-enabled and hybrid models against empirical and mechanistic baselines.
• A model-comparison report identifying where AI/hybrid approaches add value over existing approaches and where they do not.
• A final presentation summarizing project objectives, methods, results, limitations, and recommendations for next steps, with external publications.