Learning a Markov Logic network for supervised gene regulatory network inference.

Area of Science:

• Systems Biology
• Computational Biology
• Bioinformatics

Background:

• Gene regulatory network inference is a complex challenge in systems biology.
• Supervised network inference is suitable when prior knowledge of gene regulatory networks is available.
• Markov Logic Networks (MLN) integrate probabilistic graphical models with first-order logic for network inference.

Purpose of the Study:

• To develop and evaluate a supervised gene regulatory network inference method using Markov Logic Networks (MLN).
• To leverage existing biological knowledge and transcriptomic data for predicting gene regulations.
• To assess the performance of MLN in predicting novel gene-gene interactions.

Main Methods:

• Learning a Markov Logic Network (MLN) model from a known gene regulatory network and transcriptomic data.
• Encoding genes and their relationships into first-order logic for MLN input.
• Employing asymmetric bagging to handle unbalanced training data for MLN learning.
• Developing in silico tests to evaluate pairwise classifiers for network inference tasks.

Main Results:

• MLN, when incorporating heterogeneous gene property descriptions, achieves performance comparable to Support Vector Machines (SVM).
• MLN demonstrates strong predictive performance, offering insights into inferred gene regulations.
• The proposed in silico tests provide a framework for assessing classifier performance in network inference.

Conclusions:

• MLN provides a powerful and interpretable approach for gene regulatory network inference.
• The method facilitates the discovery of new gene regulations and cross-validation of experimental findings.
• MLN integrates diverse biological data sources for enhanced network prediction.