Multi-omics regulatory network inference in the presence of missing data

Area of Science:

• Systems Biology
• Bioinformatics
• Genomics

Background:

• Systems biology aims to uncover regulatory mechanisms driving complex biological systems using multi-level networks.
• Multi-omics profiling is crucial for understanding these networks but often suffers from missing data due to experimental constraints.
• Classical computational methods for regulatory network inference are limited when dealing with missing data.

Purpose of the Study:

• To integrate regression-based methods capable of handling missing data into the KiMONo (Knowledge guided Multi-Omics Network inference) approach.
• To benchmark the performance of these integrated methods in various missing data scenarios common in single- and multi-omics studies.
• To assess the feasibility of robust multi-omics network inference despite the presence of missing data.

Main Methods:

• Integration of regression methods designed to handle missing data into the KiMONo framework.
• Benchmarking performance across diverse missing data scenarios, including random and block missingness.
• Evaluation on both single- and multi-omics datasets with varying omics-layer dimensions.

Main Results:

• Two-step approaches explicitly handling missingness demonstrated superior performance in random and block missingness scenarios on imbalanced omics-layer dimensions.
• Methods implicitly handling missingness showed optimal performance on balanced omics-layer dimensions.
• KiMONo successfully enabled robust multi-omics network inference even with missing data.

Conclusions:

• Robust multi-omics network inference in the presence of missing data using KiMONo is feasible.
• The developed approach allows for the full utilization of available multi-omics data, overcoming limitations of missing information.
• This facilitates a more comprehensive understanding of regulatory mechanisms in complex biological systems.