Precisely identify the m6A sites from miCLIP data is still a challenge in the epigenomics field. Here we present a workflow to determine the m6A sites from the miCLIP2 data set.
N6-methyladenosine (m6A) is the most abundant internal modification in mRNA. It impacts many different aspects of an mRNA’s life, e.g. nuclear export, translation, stability, etc.
m6A individual-nucleotide resolution UV crosslinking and immunoprecipitation (miCLIP) and the improved miCLIP2 are m6A antibody-based methods that allow the transcriptome-wide mapping of m6A sites at a single-nucleotide resolution (Körtel et al. 2021)(Linder et al. 2015). In brief, UV crosslinking of the m6A antibody to the modified RNA leads to truncation of reverse transcription or C-to-T transitions in the case of readthrough. However, due to the limited specificity and high cross-reactivity of the m6A antibodies, the miCLIP data comprise a high background signal, which hampers the reliable identification of m6A sites from the data.
For accurately detecting m6A sites, we implemented an AdaBoost-based machine learning model (m6Aboost) for classifying the miCLIP2 peaks into m6A sites and background signals (Körtel et al. 2021). The model was trained on high-confidence m6A sites that were obtained by comparing wildtype and Mettl3 knockout mouse embryonic stem cells (mESC) lacking the major methyltransferase Mettl3. For classification, the m6Aboost model uses a series of features, including the experimental miCLIP2 signal (truncation events and C-to-T transitions) as well as the transcript region (5’UTR, CDS, 3’UTR) and the nucleotide sequence in a 21-nt window around the miCLIP2 peak.
The package m6Aboost includes the trained model and the functionalities to prepare the data, extract the required features and predict the m6A sites.
if (!require("BiocManager")) install.packages("BiocManager") BiocManager::install("m6Aboost")
A package only needs to be installed once. Load the package into an R session with
The workflow described herein is based on our published paper (Körtel et al. 2021). Thus we expect the user to preprocess the miCLIP2 data based on the preprocessing pipeline in our article (Körtel et al. 2021). In brief, the preprocessing steps include basic processing of the sequencing reads, such as quality filtering, barcode handling, mapping, generation of the single nucleotide crosslink and the C to T transition bigWig file. After the preprocessing, we expect the user to do the peak calling with the tool PureCLIP (Krakau, Richard, and Marsico 2017).
Note: If you use
m6Aboost in published research, please cite:
Körtel, Nadine#, Cornelia Rückle#, You Zhou#, Anke Busch, Peter Hoch-Kraft, FX Reymond Sutandy, Jacob Haase, et al. 2021. “Deep and accurate detection of m6A RNA modifications using miCLIP2 and m6Aboost machine learning.” bioRxiv. https://doi.org/10.1101/2020.12.20.423675.