SuMO-Fil: Supervised multi-omic filtering prior to performing network analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• Multi-omic analyses integrate high-dimensional datasets, often facing statistical power limitations and computational burdens.
• Existing methods struggle with noise and irrelevant variables in sparse settings, hindering downstream analyses like network detection.

Purpose of the Study:

• To introduce SuMO-Fil, a pre-processing method for Supervised Multi-Omic Filtering, designed to address computational and statistical power deficiencies.
• To improve the accuracy and reduce the runtime of downstream analyses, particularly those involving sparse gene network detection.

Main Methods:

• SuMO-Fil implements variable filtering based on low inter-dataset similarity and low similarity with the outcome variable.
• The method was evaluated by simulating modular networks and comparing its performance against traditional low-mean or low-variance filtering techniques.
• Specific filtering methods for cluster and network analysis were developed and assessed.

Main Results:

• SuMO-Fil effectively eliminates non-network features while preserving crucial biological signals across various settings.
• Compared to existing filtering techniques, SuMO-Fil demonstrates superior performance in maintaining biological relevance.
• The speed and accuracy of downstream methods, such as supervised sparse canonical correlation analysis, significantly improved after applying SuMO-Fil.

Conclusions:

• SuMO-Fil enhances the scalability and efficiency of multi-omic data analysis by reducing computational demands and improving statistical power.
• This pre-processing approach is particularly beneficial for supervised network inference in sparse, high-dimensional omics data.
• SuMO-Fil offers a robust solution for noise reduction and signal preservation in complex biological datasets.