PhyteByte

PhyteByte predicts bioactive interactions between food-derived compounds and specified protein targets using machine learning on molecular structures.

Key Features:

• Chemoinformatic Approach: Employs chemoinformatic methods to infer compound bioactivity from molecular structures.
• Structure-Based Activity Prediction: Uses structure-based activity prediction leveraging pharmacological structure and activity data from ChEMBL.
• Random Forest Classifier: Trains a random forest classifier for each input protein target using molecular fingerprints and ChEMBL-derived labels.
• Integration with FooDB Database: Queries the FooDB library of food compounds providing molecular structures, source foods, and quantity information where available.
• Output Generation: Produces ranked lists of food compounds predicted to have biological effects with associated source foods and available quantity data.

Scientific Applications:

• Targeted compound identification: Identify food-based compounds that may interact with specific protein targets such as PPARG.
• Candidate prioritization: Prioritize potential agonists including irigenin, sesamin, fargesin, and delta-sanshool for experimental or epidemiological follow-up.
• Nutraceutical and dietary research: Support selection of compounds for precision nutraceutical development and personalized dietary investigations.

Methodology:

Input a target protein, retrieve structure and activity data from ChEMBL, compute molecular fingerprints for food compounds from FooDB, train a random forest classifier per target on ChEMBL labels, and apply the classifier to FooDB compounds to generate predicted bioactive compound lists with source and quantity annotations.