NanoCNN: A CNN-Based Tool to Detect RNA m ⁶ A Methylations Using Oxford Nanopore Sequencing

We introduce a convolutional neural network (CNN)-based tool named NanoCNN, designed specifically for detecting m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">6</sup>A-modified sites in Oxford Nanopore sequencing signals. As a significant regulator of gene expression, m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">6</sup>A plays a crucial role in cellular processes, and direct RNA sequencing utilizing nanopore signals offers a promising avenue for its study. While existing computational tools aim to address this need, they often rely solely on current or alignment features, resulting in suboptimal performance. In contrast, our approach employs a signal-processing method that extracts features such as power spectra, root mean square (RMS), kurtosis, moving average, and crest factors from each Adenine nucleotide base obtained from publicly available nanopore signal datasets. We utilized over 348 million such bases for training NanoCNN and over 40 million bases to evaluate its performance. Additionally, we benchmarked the performance metrics, specifically the area under the curve (AUC), of NanoCNN against leading tools such as Epinano, xPore, and CHEUI for site-based classification. Our NanoCNN achieved an impressive AUC of 0.8867, surpassing the performance of these tools, albeit at the cost of longer durations for feature extraction and classification. Notably, NanoCNN is the first tool of its kind to adopt a signal-processing approach for detecting RNA m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">6</sup>A-modified bases using Oxford Nanopore signals.