NmSEER: A Prediction Tool for 2'-O-Methylation (Nm) Sites Based on Random Forest

摘要

2'-O-methylation (2'-O-me or Nm) is a common RNA modification, which was initially discovered in various non-coding RNAs. Recent researches also revealed its prevalence and regulatory importance in mRNA. In this work, we first demonstrate that the Nm sites can be accurately predicted by the RNA sequence features. By utilizing simple one-hot encoding scheme of positional nucleotide sequence and the random forest machine learning algorithm, we developed a computational prediction tool named NmSEER to predict Nm sites in HeLa cells, HEK293 cells or both of them. Based on our observation of the subgrouping of the Nm sites, we proposed a specialized subgroup-wise prediction strategy to further enhance the prediction performance for the Nm sites with the consensus AGAT motif. Our predictor has achieved a promising performance in both the cross-validation test and the independent test (AUROC = 0.909 and 0.928 for predicting AGAT-sites and non-AGAT sites in independent test, respectively). NmSEER is implemented as a user-friendly web server, which is freely available at.