Weakly supervised learning of RNA modifications from low-resolution epitranscriptome data

Abstract

Motivation Increasing evidence suggests that post-transcriptional ribonucleic acid (RNA) modifications regulate essential biomolecular functions and are related to the pathogenesis of various diseases. Precise identification of RNA modification sites is essential for understanding the regulatory mechanisms of RNAs. To date, many computational approaches for predicting RNA modifications have been developed, most of which were based on strong supervision enabled by base-resolution epitranscriptome data. However, high-resolution data may not be available. Results We propose WeakRM, the first weakly supervised learning framework for predicting RNA modifications from low-resolution epitranscriptome datasets, such as those generated from acRIP-seq and hMeRIP-seq. Evaluations on three independent datasets (corresponding to three different RNA modification types and their respective sequencing technologies) demonstrated the effectiveness of our approach in predicting RNA modifications from low-resolution data. WeakRM outperformed state-of-the-art multi-instance learning methods for genomic sequences, such as WSCNN, which was originally designed for transcription factor binding site prediction. Additionally, our approach captured motifs that are consistent with existing knowledge, and visualization of the predicted modification-containing regions unveiled the potentials of detecting RNA modifications with improved resolution. Availability implementation The source code for the WeakRM algorithm, along with the datasets used, are freely accessible at: https://github.com/daiyun02211/WeakRM Supplementary information Supplementary data are available at Bioinformatics online.

i5hmCVec: Identifying 5-Hydroxymethylcytosine Sites of Drosophila RNA Using Sequence Feature Embeddings

“…The i5hmCVec is constructed based on a low-resolution modification dataset. WeakRM ( Huang et al, 2021 ) was also proposed for identifying the 5hmC modification sites on low-resolution data. We summarized the dataset distribution used in the i5hmCVec and WeakRM in Table 1 .…”

M1ARegpred: Epitranscriptome Target Prediction of N1-methyladenosine (m1A) Regulators Based on Sequencing Features and Genomic Features

“…Secondly, the performance can be further improved. Recent studies applying the deep learning algorithms showed effectiveness in site predictions [64][65][66]. Therefore, increasing the current genome features or applying the deep learning algorithm may contribute to better performance.…”

Geographic encoding of transcripts enabled high-accuracy and isoform-aware deep learning of RNA methylation

“…As for tissue-specific MeRIP-seq based m 6 A data, only coarse-grained labels are available, which means that we only know whether a peak (genome bin) contains m 6 A sites or not, but we do not know which adenosine is modifiable. We previously developed WeakRM ( 45 ), a weakly supervised learning framework that takes genome bin data of various widths as input and learns context-specific RNA methylation patterns. In our tissue-specific m 6 A prediction problem, the instance length was set to 50, and the instance stride was set to 10.…”

“…Since experimental approaches for studying RNA modification are expensive and laborious, in silico methods have drawn increasing attention as an alternative avenue and have achieved great success in recent years. To date, more than 100 different approaches ( 22–26 ) have been established for computational prediction of RNA modification sites, including most notably, the iRNA toolkit ( 27–36 ), SRAMP ( 37 ), WHISTLE ( 38 ), Gene2vec ( 39 ), PEA ( 40 ), DeepPromise ( 25 ), MASS ( 41 ), m6Aboost ( 42 ), MultiRM ( 43 ), DeepAc4C ( 44 ), WeakRM ( 45 ), PULSE ( 46 ), NmRF ( 47 ), etc. Among them, the iRNA toolkit ( 27–36 ) developed primarily by Chen, Lin and Chou is the earliest as well as the most versatile toolkit, supporting multiple RNA modification types based on RNA primary sequences and has been widely recognized as the gold standard for benchmarking the accuracy of different RNA modification prediction approaches.…”

“…Additionally, we explored the impact of isoform ambiguity on m 6 A site prediction and developed an attention-based multiple instance learning framework to fully use the isoform transcript information. By combining our previously developed WeakRM framework ( 45 ) and Geo2vec, we constructed isoform-aware high-accuracy tissue-specific m 6 A predictors for 25 human tissues (with mean AUC of 0.893 and mean AP 0.873). Compared with the sequence-only model, the AUC is 8% higher, and the AP is 10.7% higher, showing the importance of distinguishing isoform-specific methylation.…”