Hierarchical DNN with Heterogeneous Computing Enabled High-Performance DNA Sequencing

Abstract

DNA sequencing is a popular tool to demystify the code of living organisms and is reforming the medical, pharmaceutical and biotech industries. The Next-Generation Sequencing (NGS) plays a vital role in high-throughput DNA sequencing with massively parallel data generation. Nevertheless, the massive amount of data imposes great challenges for data analysis. It is arduous to reach a low error rate for handling noisy and/or biased signals owing to the imperfect biochemical reactions and imaging systems. Furthermore, a homogeneous computing system lacks computing power and memory bandwidth. Therefore, in this work, a heterogeneous computing platform with a hierarchical deep neural network sequencing pipeline is proposed to improve the sequencing quality and increase processing speed. Experiments demonstrate that the proposed work reached higher effective throughput (12.18% more clusters found), lower error rate (0.0175%), higher quality score (%Q30 99.27%), and 19% faster. The reported work empowers virus detection, diseases diagnostic, and other potential biomedical applications.