Intrinsic-dimension analysis for guiding dimensionality reduction and data fusion in multi-omics data processing.

Area of Science:

• Biomedical Research
• Computational Biology
• Genomics

Background:

• Multi-omics data offers comprehensive biological insights but presents analytical challenges.
• High dimensionality and limited sample sizes necessitate effective data reduction and integration.
• Existing methods often overlook unique challenges of individual omics data.

Purpose of the Study:

• To develop a novel multi-modal dimensionality reduction pipeline for multi-omics data.
• To address the limitations of uniform dimensionality reduction across different omics.
• To improve the reliability and accuracy of multi-omics data analysis.

Main Methods:

• Utilized intrinsic dimensionality estimators to assess the curse-of-dimensionality impact.
• Proposed a two-step reduction strategy combining feature selection and extraction for affected views.
• Explored three unsupervised multi-omics data fusion methods.

Main Results:

• The novel pipeline demonstrated significant improvements over traditional uniform reduction methods.
• The tailored approach enhances supervised multi-omics analysis.
• Insights into the performance of unsupervised fusion methods were gained.

Conclusions:

• A novel, view-specific dimensionality reduction pipeline significantly improves multi-omics data analysis.
• This approach offers a more robust framework for understanding biological systems and disease mechanisms.
• Further exploration of unsupervised fusion methods in critical settings is warranted.