Data and code for benchmarking siRNA efficacy predictors.
- data/: Download and prepare the dataset, features and splits.
- benchmark/: Run the published predictors and baselines, and evaluate predictions.
- models/: Train and run the five Agentomics models and TabPFN.
The published predictors use Docker. Data scripts, Agentomics and TabPFN use uv.
Datasets, training outputs, predictions, fitted models and a code snapshot are in Zenodo version 3. The original experimental snapshot remains in version 2, used by the data and evaluation scripts.
First-party code is MIT licensed. External implementations, model weights and source datasets retain their original terms. TabPFN weights, OligoFormer source and Rosetta are obtained separately under their respective licences.