Prediction of plant secondary metabolic pathways using deep transfer learning.

Area of Science:

• Biochemistry
• Computational Biology
• Cheminformatics

Background:

• Plant secondary metabolites are crucial for pharmaceuticals, nutrition, and other industries.
• Elucidating plant metabolic pathways is vital for understanding plant biology.
• Current databases lack sufficient information on plant biosynthesis and degradation pathways.

Purpose of the Study:

• To develop a novel deep learning approach for predicting plant metabolic pathways.
• To address the challenge of incomplete information in current plant pathway databases.
• To leverage transfer learning for enhanced prediction accuracy.

Main Methods:

• A hybrid deep learning architecture combining Graph Transformer and convolutional neural network (GTC) was developed.
• The GTC model was pre-trained on KEGG datasets and fine-tuned on plant-derived datasets.
• Transfer learning was employed to transfer knowledge from general metabolic pathways to plant-specific ones.

Main Results:

• The GTC model achieved high performance on the KEGG dataset, outperforming six other models.
• GTC demonstrated excellent accuracy in predicting plant secondary metabolic pathways (average 98.30%).
• The model achieved perfect accuracy (100.00%) in classifying alkaloids and high accuracy for other natural products.

Conclusions:

• The GTC model effectively captures molecular features for pathway prediction and natural product classification.
• Transfer learning significantly enhances the prediction of plant secondary metabolic pathways.
• A user-friendly program is available for compound analysis using SMILES strings.