Model Gateway: Management platform for model-driven drug discovery

Area of Science:

• Drug discovery and development
• Computational chemistry
• Machine learning in pharmaceuticals

Background:

• Pharmaceutical drug discovery requires specialized machine learning (ML) infrastructure beyond general ML Operations.
• Existing platforms lack comprehensive solutions for multiparameter optimization, physics-based model versioning, and pharma-specific compound library governance.
• The unique demands include inference-time model composition and scientific organizational unit-based controls.

Purpose of the Study:

• To present the Model Gateway, a cloud-based platform designed to manage ML and scientific computational models within drug discovery pipelines.
• To address the limitations of current platforms by providing a unified solution for complex pharmaceutical R&D needs.
• To facilitate efficient and structured management of diverse computational models across the drug discovery workflow.

Main Methods:

• The Model Gateway platform offers centralized version control for ML and scientific models.
• It implements pharma-structured governance, asynchronous execution, and consensus model orchestration.
• Key features include automated retraining and a unified API for various clients like molecular design suites and large language model agents.

Main Results:

• The platform is in production at Eli Lilly, managing over 200 deployed models.
• These models span small-molecule, peptide, and antibody modalities.
• Model Gateway serves more than five downstream applications across all phases of the Design-Make-Test-Analyze cycle.

Conclusions:

• Model Gateway provides a comprehensive solution for managing diverse computational models in pharmaceutical R&D.
• It enhances the efficiency and structure of the drug discovery pipeline through advanced ML infrastructure.
• The platform's successful implementation demonstrates its value in governing and deploying a large number of scientific models.