Multiple Linear Regression for Reconstruction of Gene Regulatory Networks in Solving Cascade Error Problems

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Gene regulatory network (GRN) reconstruction is crucial for understanding gene interactions.
• Previous GRN methods often suffer from inaccurate prediction of cascade motifs, mistaking indirect interactions for direct ones.
• A lack of specific methods addressing cascade errors hinders GRN prediction performance.

Purpose of the Study:

• To propose Multiple Linear Regression (MLR) as a method for inferring GRNs from gene expression data.
• To specifically address and avoid the misinterpretation of indirect interactions (A → B → C) as direct interactions (A → C) in GRN reconstruction.
• To evaluate the effectiveness of MLR in minimizing cascade errors using a novel experimental procedure.

Main Methods:

• Utilized Multiple Linear Regression (MLR) for GRN inference from gene expression data.
• Employed random subnetwork extraction to manage datasets with fewer observations than predictors.
• Developed and applied a novel experimental procedure to specifically assess MLR's performance in avoiding cascade errors.

Main Results:

• The proposed MLR method resulted in a minimal number of cascade errors.
• The Belsley collinearity test indicated significant multicollinearity issues in the experimental datasets.
• All tested subnetworks achieved satisfactory performance, with Area Under the Receiver Operating Characteristic (AUROC) values exceeding 0.5.

Conclusions:

• Multiple Linear Regression (MLR) is an effective method for gene regulatory network (GRN) reconstruction, particularly in mitigating cascade errors.
• The novel experimental procedure validated MLR's capability to accurately distinguish direct and indirect gene interactions.
• Despite challenges like multicollinearity, MLR demonstrates robust performance in GRN inference from gene expression data.