HGAlign: Biologically preserving batch correction and classification for metabolomics via heterogeneous graph...

Area of Science:

• Computational Biology
• Analytical Chemistry
• Biotechnology

Background:

• Large-scale metabolomics using Matrix-Assisted Laser Desorption/Ionization Mass Spectrometry (MALDI-MS) is crucial for biological profiling.
• Batch effects in MALDI-MS experiments, stemming from processing and instrument variations, obscure biological signals and reduce reproducibility.
• Current batch correction methods often oversimplify or risk removing vital biological information.

Purpose of the Study:

• To develop a robust method for correcting batch effects in large-scale MALDI-MS data.
• To preserve biologically relevant information, such as disease-specific metabolic differences, during batch correction.
• To improve the accuracy and generalizability of downstream analyses in metabolomics.

Main Methods:

• Introduction of HGAlign (Heterogeneous Graph Alignment Model), a neural network utilizing heterogeneous graph convolutional networks.
• Learning relationships between samples and metabolic features to enable effective batch correction.
• Application to CyTOF public datasets and clinical MALDI-MS serum data from systemic lupus erythematosus (SLE) patients.

Main Results:

• HGAlign significantly reduces inter-batch discrepancies in MALDI-MS data.
• The model maintains or improves classification accuracy for disease prediction (e.g., SLE).
• HGAlign achieves state-of-the-art performance, demonstrated by the lowest Minimum Molecular Distance (MMD) values and improved classification metrics, while avoiding over-correction.

Conclusions:

• HGAlign provides a principled framework for balancing batch effect removal and biological signal preservation in high-throughput metabolomics.
• Heterogeneous graph representation learning enhances batch correction and disease classification performance.
• HGAlign shows strong potential for large-scale clinical applications in metabolomics research.