Identification of differentially methylated single-nucleotide m6A sites by incorporating site-specific antibody specificity

Various genome-wide and transcriptome-wide technologies are based on antibodies, however, the specificity of antibodies on different targets has not been characterized or considered in the analyses. The antibody-based MeRIP-seq is the most widely used method to determine the locations of N6-methyladenosine (m6A) on RNAs, especially for differential m6A analyses. However, the antibody specificities in different RNA regions and their resulting technical biases in differential m6A analyses have not been evaluated. Here, we evaluated the m6A antibody specificities using 100 pairs of spike-in RNAs with known m6A levels at single sites. Based on two replicates with different m6A levels on spike-in RNAs, we realized the m6A antibody specificities of the m6A sites on spike-in RNAs were greatly varied and mainly determined by the surrounding sequences of the m6A sites. Moreover, the MeRIP-seq signal fold change is the function of the real difference in m6A levels as well as the m6A antibody specificity. We then trained a machine learning model to predict the m6A antibody specificities of given sequences and predicted the m6A specificities of all RNA sequences surrounding the known m6A motif DRACH throughout the human transcriptome. Finally, we developed a Hierarchical statistic model for Differential Analysis of m6A Sites (HDAMS) by taking advantage of the predicted m6A specificities. We found that HDAMS can accurately determine the differentially methylated single-nucleotide m6A sites and the output more functionally relevant results. Our study not only provides a powerful tool for differential m6A analyses but also provides a methodological framework for other antibody-based studies to incorporate antibody specificities.