ARBic: An All-Round Biclustering Algorithm for Analyzing Gene Expression Data
Abstract Background: Identifying significant biclusters of genes with specific expression patterns is an effective approach to reveal functionally correlated genes in gene expression data. However, existing algorithms are limited to finding either broad or narrow biclusters but both due to failure of balancing between effectiveness and efficiency. Methods: We developed a new algorithm ARBic which can accurately identify any meaningful biclusters of shape no matter broad or narrow in a large scale gene expression data matrix, even when the values in the biclusters to be identified have the same distribution as that the background data has. ARBic is developed by integrating column-based and row-based strategies into biclustering procedure. The column-based strategy borrowed from ReBic, a recently published biclustering tool, prefers to narrow bicluters. The row-based strategy newly designed in this article by repeatedly finding a longest path in a specific directed graph prefers to broader ones. Result and Conclusion: When tested and compared to other seven salient biclustering algorithms on simulated datasets, ARBic achieved recovery, relevance and f1-scores 29% higher than the second best algorithm. Furthermore, ARBic substantially outperforms all of them on real datasets and robusts to noises, shapes of biclusters and types of datasets.Code: https://github.com/holyzews/ARBicData: https://doi.org/10.5281/zenodo.5121018