Integrating Sunflower Optimization with Recursive Feature Elimination for Transcriptomic Biomarker Recognition

The precise identification of biomarkers is a key link in promoting the development of precision medicine. Omics data are inherently high-dimensional but often limited in sample size, where abundant redundant and noisy features significantly impede biomarker discovery. The feature selection algorithm can effectively mine the key information in the data without changing the physical properties of the original features, and thus has become an important tool for identifying object markers in groups. This study proposes a novel feature selection algorithm - sunflower optimization recursive feature elimination algorithm (SFO-RFE). This algorithm constructs a feature selection framework by integrating the global search ability of the sunflower optimization algorithm (SFO) and the iterative screening mechanism of recursive feature elimination (RFE). Firstly, a preliminary feature screening is conducted based on the t-test. Then, the improved sunflower optimization algorithm is combined to optimize the feature subset. For the poorly performing search agents, Brownian motion is introduced to enhance the global search ability. Combined with RFE algorithm, redundant features are dynamically eliminated every 10 iterations, thereby constructing a prediction model with high classification performance. Experiments on five independent transcriptome datasets show that, compared with eight mainstream feature selection algorithms, SFO-RFE exhibits significant advantages in terms of classification accuracy, stability and the interpretability of biomarkers. This study provides a new idea for the feature selection of omics data and has potential clinical application value.