Identification of differential RNA modifications from nanopore direct RNA sequencing with xPore

Detection of m6A from direct RNA sequencing using a Multiple Instance Learning framework

“…While supervised methods can identify RNA modifications in a single sample, comparative methods facilitate the analysis across conditions 40,42,53 . However, one of the key advantages of supervised methods over comparative methods is their ability to predict the occurrence of specific RNA modifications such as m6A.…”

“…In order to evaluate whether the novel sites predicted by m6Anet are valid m6A sites, we identified positions which are sensitive to loss of the m6A writer METTL3 . Using an existing comparative approach (xPore), we mapped m6A sites in the HEK293T cell line by comparing it against a METTL3 knockout cell line that is depleted of m6A 28,42 . We then define DRACH sites which have a significant difference compared to this control as knockout sensitive sites (KO sensitive), resulting in 1888 candidate positions when a stringent threshold is used (Suppl.…”

“…As m6Anet integrates read level probabilities to obtain the final site level probability, these observations suggest that it might reflect the underlying modification stoichiometry. To validate whether a change in the proportion of modified reads is reflected in a change in site-level m6A probabilities, we analysed direct RNA-Seq data from METTL3 knockout and wild type samples that were mixed at specific proportions corresponding to an expected relative m6A stoichiometry of 0%, 25%, 50%, 75%, and 100% 42 . On the set of sites which were predicted to be modified in the 100% wild type samples (p>=0.9) and which are predicted to be unmodified in the knockout samples (p<=0.2), we observed a gradual shift of reads from the modified cluster to the unmodified cluster, corresponding to the expected changes in the relative m6A stoichiometry (Figure 4d-f, Suppl.…”

“…Processing of direct RNA sequencing data All data used in this work was obtained from 42 , 50 . To train and validate m6Anet, we downloaded a single replicate (replicate 2 run 1) of the HCT116 cell line and a single replicate of the HEK293T cell line (replicate 1) while to run xPore, we downloaded all replicates of the HEK293T cell lines as recommended .…”

“…Comparative approaches do not require training data for known RNA modifications but instead use control or reference samples to detect meaningful shifts in signal-based features that correlate to the presence of modifications. Comparative methods such as Tombo 37 , DRUMMER 38 , nanoDOC 39 , Nanocompore 40 , ELIGOS 41 , xPore 42 , and Yanocomp 43 detect m6A sites by comparing with a sample with few or no m6A modifications. While these methods are accurate, their success relies on the availability of m6A-free control samples which typically involves silencing of specific writer genes which can be a limiting factor.…”