Related Experiment Videos
SitoshnaPred: Learning phytochemical descriptors to elucidate Ayurvedic herbal potency
Ashish Panghalia1, Vikram Singh1
1Centre for Computational Biology and Bioinformatics, School of Life Sciences, Central University of Himachal Pradesh, 176215, India.
Journal of Ethnopharmacology
|September 20, 2025
Summary
This study links Ayurvedic herb potencies (Sīta/cold, Uṣṇa/hot) to their phytochemicals using machine learning. Molecular descriptors accurately predict herbal properties, validating traditional knowledge.
Area of Science:
- Phytochemistry
- Computational Biology
- Ayurvedic Medicine
Background:
- Traditional Ayurvedic medicine classifies herbs into Sīta (cold), Uṣṇa (hot), and Unuṣṇa (neutral) potencies.
- A dataset of 627 herbs with potency classifications was created from classical texts.
- This study explores the link between herbal potencies and their phytochemical constituents.
Purpose of the Study:
- To investigate the relationship between phytochemicals and the cold/hot nature of Ayurvedic herbs.
- To develop an ensemble learning model for classifying herbal potencies.
- To computationally characterize the molecular basis of traditional herbal medicine.
Main Methods:
- Developed a dataset of 454 herbs and 13,534 associated phytochemicals.
- Calculated 1613 2D and 213 3D molecular descriptors for each phytochemical.
- Utilized LightGBM for binary and ternary classification (SitoshnaPred) after dimensionality reduction.
Main Results:
- The binary SitoshnaPred model achieved 94.01% accuracy and 98.41% AUC.
- The ternary classification model achieved 79.84% accuracy and 98.10% AUC.
- SHAP and loadings analyses identified key molecular descriptors influencing classification.
Conclusions:
- Phytochemical molecular descriptors contain significant information about herbal potency.
- This approach can help characterize the molecular basis of traditional Ayurvedic knowledge.
- The findings support AI-guided prediction of herb properties.