Back

Accountable Prediction of Drug ADMET Properties with Molecular Descriptors

Boral, N.; Ghosh, P.; Goswami, A.; Bhattacharyya, M.

2022-07-02 bioinformatics
10.1101/2022.06.29.115436 bioRxiv
Show abstract

Drugs are chemical substances of low molecular weights that require to travel from the site of administration to the site of action. For safety and effective permeability, drugs are required to exhibit ideal absorption, distribution, metabolism, excretion and toxicity (ADMET) properties. Given the Simplified Molecular Input Line Entry System (SMILES) representation of drugs, we aim to predict their ADMET properties. By feeding molecular descriptors (as global features) to traditional machine learning models, we show that ADMET properties of drug molecules can be predicted with an accuracy competitive with the state-of-the-art deep learning models. We demonstrate that the proposed approach with only 31 molecular descriptors beats the state-of-the-art for 3 datasets. Moreover, it stands the second best for 2 other datasets, where none of the best provides a statistically significant improvement. We also demonstrate that two-dimensional descriptors can better represent absorption, distribution and excretion properties than the fingerprints widely used in the literature. However, they fail to distinguish metabolism and toxicity properties.

Matching journals

The top 9 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.