ersilia-os/ersilia

What is Ersilia?

Ersilia is a command‑line tool and Python library that lets you download, run, and manage pre‑trained AI/ML models for drug‑discovery‑related tasks (e.g., predicting antibiotic activity, ADMET properties, molecular representations, generative chemistry). All the models live in the Ersilia Model Hub, a curated catalog of models focused on infectious and neglected diseases.

How it works

1. Install the CLI (via Conda, pip, or directly from the repository). The CLI is the central entry point for all operations.
2. Fetch a model – each model has a short identifier (e.g., eos4e40). The CLI pulls the model’s container (Docker or Conda‑based) from the hub and stores it locally.
3. Serve the model – the tool starts a lightweight service (Docker container or Conda environment) that exposes the model’s inference API.
4. Run inference – give the service a CSV (or a list of SMILES strings) and receive predictions in a CSV/HDF5 file or a pandas DataFrame.
5. Manage the lifecycle – you can list models, view info, test them, close the service, or delete the local copy.

The same workflow is available from Python via the ersilia.api module, so scripts can fetch, serve, and run models programmatically.

Who might use it?

• Researchers in academia or industry who need quick access to validated predictive models for antimicrobial or other disease‑related compounds.
• Bioinformaticians who want to embed a model in a pipeline without dealing with complex environment setup.
• Educators / students looking for ready‑made examples of AI‑driven drug‑discovery models.

Key strengths

Aspect	Why it matters
Low‑code / CLI	No need to write boiler‑plate inference code; a single command fetches, serves, and runs a model.
Containerised models	Each model runs in its own isolated environment (Docker or Conda), avoiding dependency conflicts.
Domain‑focused catalog	All models target infectious or neglected diseases, with clear citations and provenance.
Python API	Seamless integration into notebooks or larger pipelines.
Open‑source & reproducible	GPL‑3 license, versioned releases on PyPI/conda‑forge, and a DOI for citation.

Limitations

• The hub only contains models contributed by the Ersilia community; you cannot upload arbitrary models without following their template workflow.
• Running a model requires Docker (or Conda) and enough compute resources for the underlying ML model; very large models may be slow on modest hardware.
• The CLI focuses on batch CSV input; interactive or real‑time inference would need custom scripting.

Quick start (CLI)

# 1. Install the CLI
conda create -n ersilia python=3.10 && conda activate ersilia
pip install ersilia   # or pip install -e . after cloning

# 2. Pull a model
ersilia fetch eos4e40

# 3. Start the service
ersilia serve eos4e40

# 4. Run predictions on a CSV
ersilia run -i my_input.csv -o my_output.csv

# 5. Clean up
ersilia close
ersilia delete eos4e40

Quick start (Python)

from ersilia.api import Model
m = Model('eos4e40')
m.fetch()
m.serve()
results = m.run(['CCO', 'c1ccccc1'])  # pandas DataFrame
m.close()

---

Bottom line: Ersilia provides a ready‑to‑use, low‑code interface for accessing a curated collection of AI models aimed at drug discovery for infectious diseases. It abstracts away environment management and lets scientists focus on applying the models rather than building them from scratch.