Deep learning, and bioinspired optimization algorithms for genetic marker selection and disease classification.

Abstract

AI-based cancer diagnosis and classification have emerged as a critical research field over the past decade, especially with advancements in next-generation sequencing technologies. However, omics datasets are often characterized by high dimensionality, complexity, and scalability challenges. Deep learning has been increasingly adopted to address these issues due to its strong predictive performance. Nonetheless, deep learning models remain largely black-box in nature, lacking interpretability—a crucial factor in biological contexts where the identification of biomarkers and selected features is essential for personalizing treatment protocols and guiding drug prescription. Therefore, various feature selection methods are highly sought to enhance interpretability. In this work, we propose a hybrid approach that combines deep learning with bioinspired feature selection techniques. This report provides an overview of recent advances in deep learning for oncology, particularly for analyzing omics data such as genomic and transcriptomic profiles. Applications in cancer diagnosis, prognosis, and therapeutic decision-making are explored, with a focus on the integration of multi-omics data for building clinical decision support systems. The results of the experiments showed that the classification accuracy was as much as 80% for single-omics models while deep neural networks and convolutional models performed better after bioinspired optimization. Enrichment analysis also confirmed the biological relevance of the selected features, affirming their use as clinically meaningful biomarkers. These findings demonstrate the effectiveness of our method in both boosting prediction performance and interpretability in cancer classification tasks.