MOTL: enhancing multi-omics matrix factorization with transfer learning.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Joint matrix factorization is a common technique for multi-omics data dimensionality reduction.
• This method's effectiveness diminishes significantly with limited sample sizes.
• A gap exists in analyzing small multi-omics datasets effectively.

Purpose of the Study:

• To introduce a novel framework, Multi-Omics Transfer Learning (MOTL), to address the limitations of analyzing small multi-omics datasets.
• To enhance the Multi-Omics Factor Analysis (MOFA) method by incorporating transfer learning principles.
• To improve the inference of latent factors for small datasets using knowledge from larger, heterogeneous datasets.

Main Methods:

• Developed MOTL, a transfer learning framework building upon MOFA.
• Inferred latent factors for small multi-omics target datasets using a large heterogeneous learning dataset.
• Evaluated MOTL using simulated and real-world data protocols, including glioblastoma samples.

Main Results:

• MOTL demonstrated improved factorization of multi-omics datasets with limited samples compared to standard factorization.
• The framework successfully enhanced the delineation of cancer status and subtype in glioblastoma samples.
• Transfer learning significantly boosted the performance of latent factor inference in low-sample scenarios.

Conclusions:

• MOTL effectively overcomes sample size limitations in multi-omics data analysis.
• The framework provides a robust method for enhancing MOFA by leveraging larger datasets.
• MOTL shows promise for improved biomarker discovery and patient stratification in complex diseases like glioblastoma.