ifeoluwafavour/eos77w8

Predicts side-effects of small molecule drugs

Side Effects

This model has been developed using GROVER, a Graph Neural Network pretrained with 10 million unlabelled molecules from Chembl and ZINC15. GROVER has then been fine-tuned to predict side-effects using the Side Effect Resource (SIDER) database compiled by MoleculeNet. GROVER predictions consistently outperformed other state-of-the-art methods for SIDER and other benchmark datasets from MoleculeNet.

Summary

• Predicts Side Effects for small molecules.
• Takes compound structures as input.
• Trained with the benchmark SIDER MoleculeNet dataset (1427 molecules).
• Results validated in-silico against baseline methods for the same dataset.
• Published in Rong et al, Advances in Neural Information Processing Systems 2020.
• Processed data can be found here.

Specifications

• Input: SMILES string (also accepts an InChIKey string or a molecule name string, and converts them to SMILES)
• Endpoint: probability of causing side effects in any of the 27 system organs defined by Meddra
• Results interpretation: 0: low - 1: high

History

1. Model was downloaded on 12.05.21 from TencentAILab
2. We duplicated task/predict.py and scripts/save_features.py from Tencent GitHub repository
3. Model was incorporated to Ersilia on 12/05/2021