Premexotac: Machine learning bitterants predictor for advancing pharmaceutical development.

Area of Science:

• Pharmacology
• Computational Chemistry
• Drug Discovery

Background:

• Bitter taste receptors (T2Rs) play roles beyond taste, presenting potential therapeutic targets.
• Traditional screening methods for bitter compounds face economic, time, and ethical limitations.
• In silico approaches offer a practical alternative for identifying bitterant ligands.

Purpose of the Study:

• To develop and validate a novel ligand-based (LB) computational model (Premexotac) for screening bitter compounds.
• To address the challenge of limited experimental data for bitterants and non-bitterants in LB classification.
• To explore new combinations of feature extraction, selection, and learning algorithms for improved bitterant prediction.

Main Methods:

• Development of Premexotac, a ligand-based bitterant screener.
• Utilized novel combinations of feature extraction and selection techniques.
• Employed machine learning algorithms for classification and validated performance using external datasets.

Main Results:

• Premexotac achieved a high F-1 score of up to 81% on external validation.
• Identified key molecular substructures from Extended Connectivity Fingerprints crucial for bitterness classification.
• Ranked physicochemical and topological descriptors, highlighting molecular branching and weight as important predictors.

Conclusions:

• Premexotac demonstrates significant potential as an efficient in silico tool for bitterant screening.
• The study provides valuable insights into molecular descriptors relevant for predicting bitterness.
• Further research can build upon these findings to enhance LB bitterness prediction models.