Enhancing Classification Accuracy and Speed with ML-DSP: A Game-Changer in Genomic Sequence Analysis

Even with a wide range of software tools at our disposal for comparing, analyzing, identifying, and categorizing genomic sequences, the complexity of the task remains, the task of classifying these sequences remains challenging. This difficulty is attributed to the vast datasets and inherent issues associated with classification. To overcome the limitations of existing alignment-centric and alignment-independent methods, a novel approach and software tool are required. In this research, we introduce ML-DSP, an innovative fusion of guided Machine Learning and processing of digital signals. This hybrid technique is developed to facilitate rapid, precise, and scalable classification of genetics across various taxonomic levels. The evaluation of ML-DSP involves the classification of 7396 complete mitochondrial genetics, resulting in an average accuracy of over 97% from kingdom to genus levels. Through a thorough comparison with classification software tools, using both small standard datasets and a larger dataset containing 4322 vertebrate mtDNA genetics, we demonstrate that machine learning digital signal processing beats the alignment-centric software MEGA7 (employing MUSCLE or CLUSTALW for alignment) into processing speed, during achieving kindred or even superior sorting accuracy across the data inventory. In contrast to the alignment-independent software FFP, Machine learning processing exhibits enhanced sorting accuracy and faster aggregate performance. Furthermore, starting experiments underscore the potential of ML-DSP in diverse datasets, achieving perfect accuracy in categorizing 4271 entire dengue virus genetic information into variants and 95.5% accuracy in classifying 4,710 bacterial genetic information into phyla.