LANDMark: an ensemble approach to the supervised selection of biomarkers in high-throughput sequencing data

A novel machine learning ensemble, LANDMark, demonstrates exceptional performance in identifying biomarkers from high-throughput sequencing data, achieving a balanced accuracy of 0.96 ± 0.06 when applied to metabarcoding datasets from Wood Buffalo National Park. Unlike traditional models, LANDMark uses oblique and non-linear decision boundaries, outperforming Random Forest and other classifiers like Linear Support Vector Machines in synthetic tests and real-world applications. Its ability to maintain high predictive accuracy without recursive feature elimination and its capacity to learn smoother non-linear decision boundaries set it apart as a robust tool for biomarker discovery in complex datasets.

Rudar, J., Porter, T. M., Wright, M., Golding, G. B., & Hajibabaei, M. (2022). LANDMark: an ensemble approach to the supervised selection of biomarkers in high-throughput sequencing data. BMC Bioinformatics, 23(1). https://doi.org/10.1186/s12859-022-04631-z