MultiPLIER: A Transfer Learning Framework for Transcriptomics Reveals Systemic Features of Rare Disease

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Individual gene expression datasets are often too small for unsupervised machine learning.
• Rare disease datasets are particularly limited, even when aggregated.
• Analyzing these small datasets is crucial for understanding disease mechanisms.

Purpose of the Study:

• To develop a method for analyzing small gene expression datasets, especially for rare diseases.
• To leverage transfer learning to extract meaningful biological patterns from limited data.
• To improve the understanding of biological processes related to rare disease severity.

Main Methods:

• Trained a pathway-level information extractor (PLIER) model on a large public data compendium.
• Applied transfer learning, termed MultiPLIER, to analyze small rare disease datasets.
• Extracted coordinated gene expression patterns using the pre-trained model.

Main Results:

• Models trained on the public compendium captured comprehensive biological factors.
• The MultiPLIER approach effectively analyzed small, rare disease gene expression datasets.
• Transferred models provided more effective descriptions of disease severity-related biological processes compared to dataset-specific models.

Conclusions:

• Transfer learning offers a powerful solution for analyzing small gene expression datasets.
• MultiPLIER enhances the utility of limited data for rare disease research.
• This method improves the biological interpretation of gene expression in rare conditions.