ersilia-os/ersilia

What it solves

Ersilia provides a unified, low-code platform to access pre-trained AI/ML models specifically designed for infectious and neglected disease research and drug discovery. It removes the technical barriers for researchers who may not be AI experts, allowing them to use complex models for antibiotic activity prediction, ADMET prediction, and generative chemistry without needing to manage complex environment setups.

How it works

The project consists of a central Command Line Interface (CLI) that manages the lifecycle of models. Users can browse a model hub, fetch a specific model using its unique identifier, serve it locally, and run predictions by passing input data (such as CSV files) to the model. To ensure compatibility and reproducibility, each model is packaged independently, typically using Docker containers or Conda environments, so that the specific dependencies of one model do not conflict with others.

Who it’s for

Researchers in laboratories, universities, and clinics—particularly those in the Global South—who need AI/ML tools for drug discovery and infectious disease research but may lack the deep technical expertise in machine learning engineering.

Highlights

• Low-code access: Simplifies the use of pre-trained models via a streamlined CLI.
• Independent packaging: Uses Docker and Conda to isolate model environments, ensuring seamless installation and execution.
• Diverse model library: Includes models for molecular representation, generative chemistry, and antibiotic activity prediction.
• Open-source mission: Focused on egalitarian access to knowledge and research outputs for the Global South.