Identification of Targeted Proteins by Jamu Formulas for Different Efficacies Using Machine Learning Approach

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Jamu herbs are traditional Indonesian medicines with potential therapeutic benefits.
• Understanding the molecular targets of Jamu compounds is crucial for drug development.
• In silico methods offer a powerful approach to predict herb-compound-protein interactions.

Purpose of the Study:

• To predict interactions between Jamu herb compounds and human proteins using data-intensive science and machine learning.
• To identify potential drug candidates from Jamu herbs based on predicted protein targets.
• To validate predicted interactions using existing scientific literature.

Main Methods:

• Collected compound, protein, and interaction data from open-access databases.
• Represented compounds using molecular fingerprints (e.g., MACCS) and proteins using numerical descriptors.
• Developed predictive models using machine learning algorithms, including Support Vector Machine and Random Forest.

Main Results:

• A Random Forest model utilizing MACCS fingerprints and amino acid composition achieved the highest accuracy.
• The best model predicted target proteins for 94 important Jamu compounds.
• Twenty-seven compounds were validated, belonging to seven distinct efficacy groups, with supporting evidence from literature.

Conclusions:

• The study identified several Jamu compounds with potential as drug candidates.
• Predicted compound efficacy and protein-disease associations suggest therapeutic applications.
• This in silico approach facilitates the selection of natural herb-based drug candidates.