Improved ALS Biomarker Discovery with SMOTE-Augmented Gene Expression Data

Abstract

The early identification of Amyotrophic Lateral Sclerosis (ALS), a progressive neurological disease, using blood-based transcriptome biomarker is gaining attention. The classification of ALS from blood transcriptomic data remains challenging due to class imbalance and high dimensionality. This extension of a previous study that utilized machine learning on the microarray dataset includes a synthetic data augmentation method employing the Synthetic Minority Over-sampling Technique (SMOTE) to improve classification accuracy. Following the use of Fisher Score, t-test, PCA, and Ant Colony Optimization for feature selection, SMOTE was employed to produce synthetic ALS samples and to imbalance the class distribution. Support Vector Machines, ensemble techniques, and k-Nearest Neighbors were used to assess the classifier's performance. The accuracy of all models improved, according to the results, with k-NN rising from 77.5% to 82% and SVM rising from 91.3% to 93%. Furthermore, a number of physiologically significant genes, such as MMP9 and SELL, appeared more noticeable after augmentation and matched known immune-related indicators in ALS. The augmentation technique improves both the predictive performance, and the biological validity of the biomarkers identified. These findings demonstrate the utility of SMOTE in enhancing transcriptomic classifiers.