N6-methyladenosine (m6A) is a chemical modification found in messenger RNA (mRNA), and it has gained attention since its discovery in the 1970s. This modification plays a crucial role in various biological processes, such as how RNA is processed and degraded, and it is linked to many diseases. Understanding m6A’s function requires detailed information about its presence across different RNA molecules, known as transcriptome-wide profiling.
Recently, Oxford Nanopore Technology Direct RNA Sequencing (DRS) has emerged as a promising method for detecting RNA modifications like m6A. DRS works by measuring changes in electrical current as RNA molecules pass through a tiny nanopore, but interpreting this current data to pinpoint m6A modifications can be quite challenging.
To address this, researchers at the University of Tokyo have developed a new computational tool called the m6A Transcriptome-wide Mapper (m6ATM). This tool uses advanced artificial intelligence, specifically deep neural networks, to predict where m6A modifications occur at a very detailed level (single-base resolution) using data from DRS. The m6ATM model is built with a special architecture that includes a WaveNet encoder and a unique learning method, allowing it to effectively analyze specific areas of RNA and characterize the overall pattern of m6A modifications.
Schematic diagram of m6ATM
(A) m6ATM was designed to predict m6A modifications at a given DRACH site in the reference transcriptome using DRS data. (B) Data preprocessing module for m6ATM. Raw FAST5 files were base-called, aligned, and resquiggled to generate read-level features containing signal and trace data. The yellow block represents the target interval of the five channels (signal and trace A/C/G/T) at the TGACA query site. The signal and trace values were transformed and concatenated into the final read-level feature of length 1280. (C) Architecture of WaveNet-DSMIL model for m6A prediction. The model aggregated 20–1000 reads at each site to determine whether the site is m6A-modified. (R: read-level data, S: site-level data, H: features from hidden layers).
Validation studies showed that m6ATM is highly accurate, achieving success rates between 80% and 98% when tested with various RNA samples that had different amounts of m6A modifications. It also performed better than other available tools when analyzing data from human cells. Additionally, m6ATM proved to be versatile in providing reliable information about the levels of m6A modifications and was even able to identify a specific RNA molecule, PEG10, as a potential target for m6A modification in liver cancer cells.
In summary, m6ATM represents a significant advancement in the study of m6A modifications in RNA. This tool not only enhances our understanding of RNA modifications but also opens new avenues for research into how these modifications can affect health and disease, paving the way for future developments in the field of epitranscriptomics.
Availability – Source codes of m6ATM can be obtained at https://github.com/poigit/m6ATM.