An Improved Stacked Autoencoder for Metabolomic Data Classification.

Area of Science:

• Computational biology
• Pharmacology
• Systems biology

Background:

• Traditional Mongolian medicine Naru3 (NR) exhibits high clinical efficacy for rheumatoid arthritis (RA) with minimal side effects.
• Metabolomics offers a powerful approach for traditional drug development, but faces challenges due to high-throughput, sparse, high-dimensional, and small sample data.
• Deep learning applications in metabolomic studies remain underexplored despite their potential.

Purpose of the Study:

• To develop an improved stacked autoencoder (SAE) algorithm for enhanced classification of metabolomic data.
• To evaluate the efficacy of the Hessian-free SAE (HF-SAE) algorithm in classifying metabolomic data from a Naru3-treated rheumatoid arthritis mouse model.

Main Methods:

• Established a Naru3-treated rheumatoid arthritis (RA) mouse model.
• Employed a Hessian-free SAE (HF-SAE) algorithm for metabolomic data classification, utilizing unlabeled data for pretraining and labeled data for fine-tuning with Hessian-free gradient descent optimization.
• Compared HF-SAE performance against Support Vector Machine (SVM), k-nearest neighbor (KNN), and gradient descent SAE (GD-SAE) using five-fold cross-validation.

Main Results:

• Successfully established a Naru3-treated animal model for RA research.
• The developed HF-SAE algorithm demonstrated successful classification of metabolomic data.
• The HF-SAE algorithm showed competitive or superior performance compared to SVM, KNN, and GD-SAE in classifying metabolomic data, with recorded mean square error and misclassification rates.

Conclusions:

• The improved HF-SAE algorithm is effective for classifying complex metabolomic data, facilitating traditional drug development.
• This study highlights the potential of deep learning approaches, specifically HF-SAE, in advancing metabolomic analysis for traditional medicines like Naru3.