CompareSVM: supervised, Support Vector Machine (SVM) inference of gene regularity networks.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Predicting gene regulatory networks (GRNs) from expression data is complex.
• Supervised and unsupervised methods exist, with Support Vector Machine (SVM) showing promise.
• A need exists for analyzing SVM's prediction accuracy across various kernels, conditions, and network sizes.

Purpose of the Study:

• To develop and utilize a tool (CompareSVM) for evaluating SVM kernel methods in GRN inference.
• To comprehensively analyze the prediction accuracy of different SVM kernels under varying biological experimental conditions and network sizes.

Main Methods:

• Development of CompareSVM, a tool based on SVM for GRN inference.
• Evaluation of various SVM kernel methods using simulated microarray datasets of different sizes.
• Detailed investigation of method performance across diverse experimental conditions.

Main Results:

• The accuracy of GRN inference is contingent on the experimental conditions and network size.
• CompareSVM demonstrated that SVM kernel performance varies significantly.
• Specific kernel performance was detailed for different dataset types and network scales.

Conclusions:

• For smaller networks (<200 nodes), the SVM Gaussian kernel generally outperforms other methods across knockout, knockdown, and multifactorial datasets.
• For larger networks (~500 nodes), the optimal inference method selection depends on the specific experimental condition.
• The CompareSVM tool is publicly available for further research.