Reconstruction of gene regulatory network via sparse optimization

TL;DR

This study evaluates sparse optimization algorithms for gene regulatory network inference using DREAM5 data, demonstrating that incorporating known network data improves accuracy and that voting algorithms outperform traditional methods across multiple datasets.

Contribution

The paper introduces a voting algorithm based on bagging sparse optimization methods, enhancing inference accuracy and robustness in gene regulatory network reconstruction.

Findings

Incorporating 20% known network data improves inference accuracy.

Voting algorithms outperform individual sparse optimization methods.

Our approach achieves better results than official DREAM5 benchmarks.

Abstract

In this paper, we tested several sparse optimization algorithms based on the public dataset of the DREAM5 Gene Regulatory Network Inference Challenge. And we find that introducing 20% of the regulatory network as a priori known data can provide a basis for parameter selection of inference algorithms, thus improving prediction efficiency and accuracy. In addition to testing common sparse optimization methods, we also developed voting algorithms by bagging them. Experiments on the DREAM5 dataset show that the sparse optimization-based inference of the moderation relation works well, achieving better results than the official DREAM5 results on three datasets. However, the performance of traditional independent algorithms varies greatly in the face of different datasets, while our voting algorithm achieves the best results on three of the four datasets.