Translation rate prediction and regulatory motif discovery with multi-task learning

Abstract

BackgroundMany studies have found that sequence in the 5’ untranslated regions (UTRs) impacts the translation rate of an mRNA, but the regulatory grammar that underpins this translation regulation remains elusive. Deep learning methods deployed to analyse massive sequencing datasets offer new solutions to motif discovery. However, existing works focused on extracting sequence motifs in individual datasets, which may not be generalisable to other datasets from the same cell type. We hypothesise that motifs that are genuinely involved in controlling translation rate are the ones that can be extracted from diverse datasets generated by different experimental techniques. In order to reveal more generalised cis-regulatory motifs for RNA translation, we develop a multi-task translation rate predictor, MTtrans, to integrate information from multiple datasets.ResultsCompared to single-task models, MTtrans reaches a higher prediction accuracy in all the benchmarked datasets generated by various experimental techniques. We show that features learnt in human samples are directly transferable to another dataset in yeast systems, demonstrating its robustness in identifying evolutionarily conserved sequence motifs. Furthermore, our newly generated experimental data corroborated the effect of most of the identified motifs based on MTtrans trained using multiple public datasets, further demonstrating the utility of MTtrans for discovering generalisable motifs.ConclusionsMTtrans effectively integrates biological insights from diverse experiments and allows robust extraction of translation-associated sequence motifs in 5’UTR.

Current limitations in predicting mRNA translation with deep learning models

“…The shared encoder is therefore trained on multiple data sets, while the task-specific tower is trained only on the respective task. In comparison to Optimus5' and Framepool, MTtrans provides an increase in R 2 of 0.015 − 0.06 in prediction accuracy, depending on the data set [17]). Interestingly, training MTtrans on multiple data sets at once rather than in a sequential, task-specific manner, achieved an almost similar effect.…”

“…The CNN also recovered some of the important regulatory elements such as uORFs [14]. More recently, a novel experimental design was used to accurately measure the output of yeast reporters driven by natural 5'UTRs [15], while novel DL architectures and training approaches aimed to improve prediction accuracy [16,17]. Potential limitations of DL models built from synthetic sequences is that it is a priori unclear whether the training set contains the regulatory elements that are relevant in vivo and whether the features extracted by the model generalize well across systems such as cell types and readouts of the process of interest.…”

A 5’ UTR Language Model for Decoding Untranslated Regions of mRNA and Function Predictions

“…Chen et al 9 introduced the RNA-FM embedding, which was used in conjunction with a 32-layer residual network (ResNet) to predict MRL. Zheng et al 5 developed a multi-task CNN model for TE prediction using multiple data sources.…”

“…Investigation into the role of 5' UTRs encompasses various aspects of translational control. With the growing interest in studying and designing 5' UTRs, various computational tools [5][6][7][8][9]5,7,8 have been developed to study its functions. For example, the ribosome load measures the number of ribosomes engaged in translating a given mRNA at a given time.…”

“…Additionally, RNABERT 10 and RNA-FM 9 are language models specific to RNA sequences and were shown useful for predicting the MRL. Further, 5’ UTR is also predictive of the mRNA translation efficiency (TE) 5, 8 that quantifies the rate of translation into proteins and the mRNA expression level (EL) 8 that reflects the relative abundance of the mRNA transcript in the cell. While there exist specialized machine learning models for individual tasks, there lacks a methodology to decode these functions of 5’ UTRs in a unified way.…”