Figure 1.

The computational framework of SRAMP. Two prediction modes have been built in SRAMP, i.e. the full transcript mode and the mature mRNA mode. Both prediction modes adopt the same computational framework. First, for a DRACH motif presented in the query sequence, its flanking sequence window is extracted and represented using the three sequence-based encodings. Then the encoded features will be submitted to the corresponding random forest classifiers. Each random forest classifier summarizes the output scores from 10 sub-classifiers, which were trained on all positive samples and a distinct subset of negative samples in the training dataset. Finally, the prediction scores of the random forest classifiers are combined through weighted summing formula. Four stringency thresholds correspond to the 99%, 95%, 90% and 85% specificities in 5-fold cross-validation test that are used to judge the classification and associated confidence. If analysing secondary structure function is enabled, the secondary structure context of the predicted m⁶A sites will be also provided.