A novel accelerated sparse Support Vector Machine (AS-SVM) algorithm for binary classification of DNA sequences

Since Support Vector Machines (SVM) can handle complicated patterns and high-dimensional data, they have proven to be quite effective in classifying DNA sequences. However, there are difficulties in real-world applications due to the time and space complexity of training and evaluating large-scale SVM models. In this paper, an accelerated sparse SVM (AS-SVM) is implemented on an FPGA to improve the efficiency of DNA sequence classification. The sparsity inherent in SVM models is exploited to reduce memory and computational requirements, making the process scalable and faster. The Field Programmable Gate Arrays (FPGA) based implementation leverages parallel processing capabilities to accelerate the inference phase, achieving significant performance improvements over traditional CPU-based approaches. The proposed implementation runs at a clock frequency of 132.33 MHz, utilizing only 0.86% LUTs, 0% BRAM, and 3.3% DSPs on the Xilinx Ultrascale+ ZCU102 FPGA, while consuming just 0.678 W of power. Compared to prior works, this demonstrates substantial improvements in resource efficiency, with a classification speedup of up to five orders of magnitude in comparison to CPU-based execution . The effectiveness of the proposed approach using a large-scale DNA sequence dataset, achieving high accuracy while maintaining minimal resource utilization. Unlike previous works that focus primarily on sequence alignment, the proposed research fills the gap in FPGA-based DNA sequence classification and highlights the potential of FPGA-based systems for bioinformatics applications.