dRFEtools: dynamic recursive feature elimination for omics

Abstract

Motivation: Advances in technology have generated larger omics datasets with potential applications for machine learning. In many datasets, however, cost and limited sample availability result in an excessively higher number of features as compared to observations. Moreover, biological processes are associated with networks of core and peripheral genes, while traditional feature selection approaches capture only core genes.

Results: To overcome these limitations, we present dRFEtools that implements dynamic recursive feature elimination (RFE), reducing computational time with high accuracy compared to standard RFE, expanding dynamic RFE to regression algorithms, and outputting the subsets of features that hold predictive power with and without peripheral features. dRFEtools integrates with scikit-learn (the popular Python machine learning platform) and thus provides new opportunities for dynamic RFE in large-scale omics data while enhancing its interpretability.

Availability and implementation: dRFEtools is freely available on PyPI at https://pypi.org/project/drfetools/ or on GitHub https://github.com/LieberInstitute/dRFEtools, implemented in Python 3, and supported on Linux, Windows, and Mac OS.

Figure 1.

Schematic of dynamic recursive feature elimination for dRFEtools. (A) Graphical representation of core and peripheral features (Boyle et al., 2017). (B) Flowchart showing recursive elimination process, where scikit-learn model can be either classification or regression. We ranked the dropped features and saved them for downstream analysis. (C) Feature iterator code used to generate dynamic elimination. (D) Flowchart showing the elimination steps using the developmental or out-of-bag (OOB) set. (E) Example of LOWESS fitting on dRFEtools model performance multiple classification using simulated data using area under the receiver operating characteristic curve (ROC AUC).