Post-transcriptional RNA modifications are found to play essential roles in epitranscriptome regulation on all types of RNAs. Most of existing computational methods can only be applied to the high-resolution data. However, high-resolution data may not be available because their wet-lab experiments usually require expensive cost, long experiment time and large amounts of input RNA. Here, we proposed AdaptRM, a multi-tasking computational method for integrated learning of epitranscriptomes. It was enabled by an adaptive pooling layer and several standard convolution blocks for tissue-specific and modification type-specific modification prediction. It can operate on both low-resolution and high-resolution datasets without further preprocessing input primary sequence. By formulating three case studies as an integrated multi-tasking learning problem, we trained the AdaptRM model and obtained impressive results. Our framework and web server should serve as useful tools for epitranscriptome research.
