NoiBic: a noise-tolerant biclustering algorithm for high-throughput gene expression data analysis.

Abstract

Motivation: High-throughput transcriptomic technologies generate large-scale gene expression datasets for functional module discovery. However, technical and biological noise can obscure coherent expression patterns and compromise module detection. Noise-tolerant methods are therefore needed to recover meaningful biclusters.

Results: We developed NoiBic, a noise-tolerant biclustering algorithm for high-throughput transcriptomic data. NoiBic integrates a longest approximate common subsequence-based seed identification strategy with noise-aware column expansion to improve bicluster detection under noisy conditions. Across simulated datasets with controlled settings for bicluster coherency, matrix size, bicluster number, noise level, and overlap level, NoiBic accurately recovered planted biclusters, particularly under noisy and overlapping conditions. Evaluation on bulk and single-cell RNA-seq datasets demonstrated its ability to identify biologically meaningful gene modules and cell type-associated biclusters.

Availability: NoiBic is implemented in C++17 and available at https://github.com/lcyi0208/NoiBic. The software version used in this study has been permanently archived on Zenodo at https://doi.org/10.5281/zenodo.22732377. Benchmark datasets are available at https://doi.org/10.5281/zenodo.21919392.