Enhancing prediction accuracy for high-dimensional small-sample-size microarray data cancer by combining chebyshev interpolation with new dual-net GAN

Cancer is a complex and multi-cellular interaction disease in the human immune system with compelling morbidity and mortality in patients. To improve cancer patient survival rate, medical professionals and clinicians commonly formulate prudent and effective cancer treatment plans by using the assistance of human microarray data. Studies reveal that gene expression profiling from microarray data has been frequently used to identify gene mutations in human cancers. But practically, this is an expensive and time-consuming technique when working with high-dimensional microarray cancer datasets. As a result, more recently, deep learning techniques in the field of artificial intelligence (AI) have become a widely effective and worthy technique to implement for cancer diagnosis and prognosis predictions in clinical applications. However, under high-dimensional but small-sample-size (HDSSS), statistically less sufficient patient sample cases, classic deep learning and machine learning models often fail to provide reliable prediction results in cancer diagnosis and prognosis. To tackle these tasks, this study focus effort on developing an AI approach termed Cheby-Dual-GAN virtual sample generation (VSG) technique, which combines Chebyshev interpolation with a novel dual-net GAN (Generative Adversarial Networks) generator, to generate high-dimensional artificial samples resembling the original data. We fed Chebyshev points and their membership function (MF) into the dual-net generator to generate highly representative virtual samples while avoiding mode collapse in GAN. Six microarray cancer datasets were used to demonstrate efficacy of the proposed method. Based on these microarray cancer datasets, we compared the Cheby-Dual-GAN method with two state-of-the-art VSG techniques. With different small training sample sizes, the proposed method can achieve significantly superior prediction accuracy in two predictive models compared to the two VSG techniques in terms of Accuracy (ACC), F-measure, mean absolute error (MAE), and mean absolute percentage error (MAPE) indicators. As a theoretical validation, the paired t-test is employed to determine whether the suggested method has statistically significant differences from the other methods. The results of paired t-test demonstrate that the proposed Cheby-Dual-GAN method enjoys significant improvement effects in terms of ACC, F-measure, MAE, MAPE indicators. According to our experimental results, the proposed method successfully outperforms other VSG methods for HDSSS microarray cancer datasets. • HDSSS microarray data poses an important challenge in bioinformatics. • We develop a novel Cheby-Dual-GAN VSG technique. • We fed Chebyshev interpolation and membership function into a dual-net generator. • Proposed Cheby-Dual-GAN method improves cancer prediction accuracy on microarray data. • Paired t-test elucidates significant differences among all five methods.