Next-Gen Genome Compression: Learning-Based Solutions for Reference-Free DNA Storage

With the rise of precision medicine, genomic data is being shared and analyzed at an unprecedented scale. However, the massive size of DNA sequences poses significant storage and transmission challenges. Existing genome compression techniques remain in early development compared to well-established image and video coding methods. This paper presents a machine learning-based approach for reference-free Deoxyribonucleic Acid (DNA) sequence compression. The method leverages codon-level mapping to American Standard for Information Interchange (ASCII) symbols, aligning with the probability mass function of standard English text. A neural network-based compression model is then applied to eliminate redundancy in the codon sequence. Experimental results demonstrate significant improvements in compression efficiency over existing state-of-the-art methods, particularly with PAQ-based encoding. This approach enhances genomic data accessibility, accelerates genetic marker identification, and optimizes drug discovery processes for precision medicine applications. Future research will explore deep learning advancements to further improve compression rates and computational efficiency.