Basecalling by Statistical Profiling and Hardware-Accelerated Convolutional Neural Network

Oxford Nanopore Technologies (ONT) genome sequencing technology enables the decoding of DNA and RNA sequences by monitoring electrical current fluctuations as nucleic acids pass through a protein nanopore. This work focuses on basecalling, which is the process of decoding these signals to detect a specific sequence. We explore both analytical and machine learning methods based on statistical distribution profiles of read currents per short sub-sequences, referred to as k-mers. Initially, we apply t-statistics to categorize each k-mer according to a predictive statistical model. Additionally, we investigate the use of a Convolutional Neural Network (CNN) for basecalling, where the input is an image representing the statistical profile of the raw data. This CNN model is deployed on a hardware acceleration platform to optimize energy and performance efficiency. Our findings exhibit promising accuracy, paving the way for cost-effective Nanopore-based sequencing applications.