Evolutionary Deep Learning Frameworks for Genomic Data Interpretation: A Systematic Review of Trends, Challenges, and Future Prospects

The explosion in the volume of genomic data generated at a single-molecule resolution, largely due to the development of high-throughput sequencing technologies and state-of-the-art biomedical instrumentation, has created an urgent demand for intelligent, scalable, and interpretable computational models. Classic machine learning approaches frequently suffer from high-dimensionality, noise and convoluted regulation in genomic data. Fortunately, the deep learning (DL) paradigm has come to the rescue as a transformative methodology, and evolutionary algorithms have offered potent tools for model fine-tuning, resulting in the recently-invented evolutionary deep learning (EDL). Nevertheless, despite recent interest, our understanding of how EDL approaches enable the interpretation of genomic data has been sketchy. To that end, this systematic review has aggregated and contributed to the EDL frameworks specific to genomic use cases, created from 2018 through 2025. The method complies with PRISMA recommendations and includes the literature obtained from IEEE Xplore, Scopus, Web of Science, ScienceDirect, and PubMed. The survey covers a range of architectures convolutional neural networks, recurrent neural networks, autoencoders, generative adversarial networks, graph neural networks, and transformers alongside evolutionary strategies such as genetic algorithms, particle swarm optimization, differential evolution, genetic programming, and neural architecture search. The results demonstrate large advances in motif discovery, variant effect prediction, single-cell expression analysis, genomic privacy protection and multi-omics integration. The reviewed evidence suggests that evolutionary optimization can improve selected DL workflows in genomic applications, particularly in architecture search, feature selection, hyperparameter optimization, and evolution-inspired data augmentation. However, the reported benefits vary across tasks, datasets, and evaluation settings. Nevertheless, challenges remain in computational complexity, the absence of meaningful benchmarking datasets, and low biological interpretability. In conclusion, this review highlights that EDL is an interesting and potential new direction for sophisticated genomic analysis, and outlooks for the multi-objective optimization, explainable artificial intelligence, and unified multimeric modeling are presented.