Characterization and prediction of mRNA polyadenylation sites in human genes

A Survey on Methods for Predicting Polyadenylation Sites from DNA Sequences, Bulk RNA-seq, and Single-cell RNA-seq

Alternative polyadenylation (APA) plays important roles in modulating mRNA stability, translation, and subcellular localization, and contributes extensively to shaping eukaryotic transcriptome complexity and proteome diversity. Identification of poly(A) sites (pAs) on a genome-wide scale is a critical step toward understanding the underlying mechanism of APA-mediated gene regulation. A number of established computational tools have been proposed to predict pAs from diverse genomic data. Here we provided an exhaustive overview of computational approaches for predicting pAs from DNA sequences, bulk RNA sequencing (RNA-seq) data, and single-cell RNA sequencing (scRNA-seq) data. Particularly, we examined several representative tools using bulk RNA-seq and scRNA-seq data from peripheral blood mononuclear cells and put forward operable suggestions on how to assess the reliability of pAs predicted by different tools. We also proposed practical guidelines on choosing appropriate methods applicable to diverse scenarios. Moreover, we discussed in depth the challenges in improving the performance of pA prediction and benchmarking different methods. Additionally, we highlighted outstanding challenges and opportunities using new machine learning and integrative multi-omics techniques, and provided our perspective on how computational methodologies might evolve in the future for non-3' untranslated region, tissue-specific, cross-species, and single-cell pA prediction.

From shallow to deep: some lessons learned from application of machine learning for recognition of functional genomic elements in human genome

Identification of genomic signals as indicators for functional genomic elements is one of the areas that received early and widespread application of machine learning methods. With time, the methods applied grew in variety and generally exhibited a tendency to improve their ability to identify some major genomic and transcriptomics signals. The evolution of machine learning in genomics followed a similar path to applications of machine learning in other fields. These were impacted in a major way by three dominant developments, namely an enormous increase in availability and quality of data, a significant increase in computational power available to machine learning applications, and finally, new machine learning paradigms, of which deep learning is the most well-known example. It is not easy in general to distinguish factors leading to improvements in results of applications of machine learning. This is even more so in the field of genomics, where the advent of next-generation sequencing and the increased ability to perform functional analysis of raw data have had a major effect on the applicability of machine learning in OMICS fields. In this paper, we survey the results from a subset of published work in application of machine learning in the recognition of genomic signals and regions in human genome and summarize some lessons learnt from this endeavor. There is no doubt that a significant progress has been made both in terms of accuracy and reliability of models. Questions remain however whether the progress has been sufficient and what these developments bring to the field of genomics in general and human genomics in particular. Improving usability, interpretability and accuracy of models remains an important open challenge for current and future research in application of machine learning and more generally of artificial intelligence methods in genomics.

Biomarker Identification through Multiomics Data Analysis of Prostate Cancer Prognostication Using a Deep Learning Model and Similarity Network Fusion

Simple Summary

Around 30% of men treated with adjuvant therapy experience recurrences of prostate cancer (PC). Current monitoring of the relapse of PC requires regular postoperative prostate-specific antigen (PSA) value follow-up. Our study aims to identify potential multiomics biomarkers using modern computational analytic methods, deep learning (DL), similarity network fusion (SNF), and the Cancer Genome Atlas (TCGA) prostate adenocarcinoma (PRAD) dataset. Six significantly intersected omics biomarkers from the two models, TELO2, ZMYND19, miR-143, miR-378a, cg00687383 (MED4), and cg02318866 (JMJD6; METTL23) were collected for multiomics panel construction. The difference between the Kaplan–Meier curves of high and low recurrence-risk groups generated from the multiomics panels and clinical information achieve p-value = 2.97 × 10⁻¹⁵ and C-index = 0.713, and the prediction performance of five-year recurrence achieves AUC = 0.789. The results show that the multiomics panel provided valuable biomarkers for the early detection of high-risk recurrent patients, and integrating multiomics data gave us the power to detect the complex mechanisms of cancer among the interactions of different genetic and epigenetic factors.

Abstract

This study is to identify potential multiomics biomarkers for the early detection of the prognostic recurrence of PC patients. A total of 494 prostate adenocarcinoma (PRAD) patients (60-recurrent included) from the Cancer Genome Atlas (TCGA) portal were analyzed using the autoencoder model and similarity network fusion. Then, multiomics panels were constructed according to the intersected omics biomarkers identified from the two models. Six intersected omics biomarkers, TELO2, ZMYND19, miR-143, miR-378a, cg00687383 (MED4), and cg02318866 (JMJD6; METTL23), were collected for multiomics panel construction. The difference between the Kaplan–Meier curves of high and low recurrence-risk groups generated from the multiomics panel achieved p-value = 5.33 × 10⁻⁹, which is better than the former study (p-value = 5 × 10⁻⁷). Additionally, when evaluating the selected multiomics biomarkers with clinical information (Gleason score, age, and cancer stage), a high-performance prediction model was generated with C-index = 0.713, p-value = 2.97 × 10⁻¹⁵, and AUC = 0.789. The risk score generated from the selected multiomics biomarkers worked as an effective indicator for the prediction of PRAD recurrence. This study helps us to understand the etiology and pathways of PRAD and further benefits both patients and physicians with potential prognostic biomarkers when making clinical decisions after surgical treatment.

Incorporating deep learning and multi-omics autoencoding for analysis of lung adenocarcinoma prognostication

Lung cancer is the most occurring cancer type, and its mortality rate is also the highest, among them lung adenocarcinoma (LUAD) accounts for about 40 % of lung cancer. There is an urgent need to develop a prognosis prediction model for lung adenocarcinoma. Previous LUAD prognosis studies only took single-omics data, such as mRNA or miRNA, into consideration. To this end, we proposed a deep learning-based autoencoding approach for combination of four-omics data, mRNA, miRNA, DNA methylation and copy number variations, to construct an autoencoder model, which learned representative features to differentiate the two optimal patient subgroups with a significant difference in survival (P = 4.08e-09) and good consistency index (C-index = 0.65). The multi-omics model was validated though four independent datasets, i.e. GSE81089 for mRNA (n = 198, P = 0.0083), GSE63805 for miRNA (n = 32, P = 0.018), GSE63384 for DNA methylation (n = 35, P = 0.009), and TCGA independent samples for copy number variations (n = 94, P = 0.0052). Finally, a functional analysis was performed on two survival subgroups to discover genes involved in biological processes and pathways. This is the first study incorporating deep autoencoding and four-omics data to construct a robust survival prediction model, and results show the approach is useful at predicting LUAD prognostication.

Inference of the human polyadenylation code

Motivation

Processing of transcripts at the 3′-end involves cleavage at a polyadenylation site followed by the addition of a poly(A)-tail. By selecting which site is cleaved, the process of alternative polyadenylation enables genes to produce transcript isoforms with different 3′-ends. To facilitate the identification and treatment of disease-causing mutations that affect polyadenylation and to understand the sequence determinants underlying this regulatory process, a computational model that can accurately predict polyadenylation patterns from genomic features is desirable.

Results

Previous works have focused on identifying candidate polyadenylation sites and classifying tissue-specific sites. By training on how multiple sites in genes are competitively selected for polyadenylation from 3′-end sequencing data, we developed a deep learning model that can predict the tissue-specific strength of a polyadenylation site in the 3′ untranslated region of the human genome given only its genomic sequence. We demonstrate the model’s broad utility on multiple tasks, without any application-specific training. The model can be used to predict which polyadenylation site is more likely to be selected in genes with multiple sites. It can be used to scan the 3′ untranslated region to find candidate polyadenylation sites. It can be used to classify the pathogenicity of variants near annotated polyadenylation sites in ClinVar. It can also be used to anticipate the effect of antisense oligonucleotide experiments to redirect polyadenylation. We provide analysis on how different features affect the model’s predictive performance and a method to identify sensitive regions of the genome at the single-based resolution that can affect polyadenylation regulation.

Supplementary information

Supplementary data are available at Bioinformatics online.

Sequence determinants of polyadenylation-mediated regulation

The cleavage and polyadenylation reaction is a crucial step in transcription termination and pre-mRNA maturation in human cells. Despite extensive research, the encoding of polyadenylation-mediated regulation of gene expression within the DNA sequence is not well understood. Here, we utilized a massively parallel reporter assay to inspect the effect of over 12,000 rationally designed polyadenylation sequences (PASs) on reporter gene expression and cleavage efficiency. We find that the PAS sequence can modulate gene expression by over five orders of magnitude. By using a uniquely designed scanning mutagenesis data set, we gain mechanistic insight into various modes of action by which the cleavage efficiency affects the sensitivity or robustness of the PAS to mutation. Furthermore, we employ motif discovery to identify both known and novel sequence motifs associated with PAS-mediated regulation. By leveraging the large scale of our data, we train a deep learning model for the highly accurate prediction of RNA levels from DNA sequence alone (R = 0.83). Moreover, we devise unique approaches for predicting exact cleavage sites for our reporter constructs and for endogenous transcripts. Taken together, our results expand our understanding of PAS-mediated regulation, and provide an unprecedented resource for analyzing and predicting PAS for regulatory genomics applications.

The prediction of virus mutation using neural networks and rough set techniques

Viral evolution remains to be a main obstacle in the effectiveness of antiviral treatments. The ability to predict this evolution will help in the early detection of drug-resistant strains and will potentially facilitate the design of more efficient antiviral treatments. Various tools has been utilized in genome studies to achieve this goal. One of these tools is machine learning, which facilitates the study of structure-activity relationships, secondary and tertiary structure evolution prediction, and sequence error correction. This work proposes a novel machine learning technique for the prediction of the possible point mutations that appear on alignments of primary RNA sequence structure. It predicts the genotype of each nucleotide in the RNA sequence, and proves that a nucleotide in an RNA sequence changes based on the other nucleotides in the sequence. Neural networks technique is utilized in order to predict new strains, then a rough set theory based algorithm is introduced to extract these point mutation patterns. This algorithm is applied on a number of aligned RNA isolates time-series species of the Newcastle virus. Two different data sets from two sources are used in the validation of these techniques. The results show that the accuracy of this technique in predicting the nucleotides in the new generation is as high as 75 %. The mutation rules are visualized for the analysis of the correlation between different nucleotides in the same RNA sequence.

Genome-wide identification and predictive modeling of tissue-specific alternative polyadenylation

Motivation: Pre-mRNA cleavage and polyadenylation are essential steps for 3′-end maturation and subsequent stability and degradation of mRNAs. This process is highly controlled by cis-regulatory elements surrounding the cleavage/polyadenylation sites (polyA sites), which are frequently constrained by sequence content and position. More than 50% of human transcripts have multiple functional polyA sites, and the specific use of alternative polyA sites (APA) results in isoforms with variable 3′-untranslated regions, thus potentially affecting gene regulation. Elucidating the regulatory mechanisms underlying differential polyA preferences in multiple cell types has been hindered both by the lack of suitable data on the precise location of cleavage sites, as well as of appropriate tests for determining APAs with significant differences across multiple libraries.

Results: We applied a tailored paired-end RNA-seq protocol to specifically probe the position of polyA sites in three human adult tissue types. We specified a linear-effects regression model to identify tissue-specific biases indicating regulated APA; the significance of differences between tissue types was assessed by an appropriately designed permutation test. This combination allowed to identify highly specific subsets of APA events in the individual tissue types. Predictive models successfully classified constitutive polyA sites from a biologically relevant background (auROC = 99.6%), as well as tissue-specific regulated sets from each other. We found that the main cis-regulatory elements described for polyadenylation are a strong, and highly informative, hallmark for constitutive sites only. Tissue-specific regulated sites were found to contain other regulatory motifs, with the canonical polyadenylation signal being nearly absent at brain-specific polyA sites. Together, our results contribute to the understanding of the diversity of post-transcriptional gene regulation.

Availability: Raw data are deposited on SRA, accession numbers: brain SRX208132, kidney SRX208087 and liver SRX208134. Processed datasets as well as model code are published on our website: http://www.genome.duke.edu/labs/ohler/research/UTR/

Contact:uwe.ohler@duke.edu

Poly(A) motif prediction using spectral latent features from human DNA sequences

MOTIVATION

Polyadenylation is the addition of a poly(A) tail to an RNA molecule. Identifying DNA sequence motifs that signal the addition of poly(A) tails is essential to improved genome annotation and better understanding of the regulatory mechanisms and stability of mRNA. Existing poly(A) motif predictors demonstrate that information extracted from the surrounding nucleotide sequences of candidate poly(A) motifs can differentiate true motifs from the false ones to a great extent. A variety of sophisticated features has been explored, including sequential, structural, statistical, thermodynamic and evolutionary properties. However, most of these methods involve extensive manual feature engineering, which can be time-consuming and can require in-depth domain knowledge.

RESULTS

We propose a novel machine-learning method for poly(A) motif prediction by marrying generative learning (hidden Markov models) and discriminative learning (support vector machines). Generative learning provides a rich palette on which the uncertainty and diversity of sequence information can be handled, while discriminative learning allows the performance of the classification task to be directly optimized. Here, we used hidden Markov models for fitting the DNA sequence dynamics, and developed an efficient spectral algorithm for extracting latent variable information from these models. These spectral latent features were then fed into support vector machines to fine-tune the classification performance. We evaluated our proposed method on a comprehensive human poly(A) dataset that consists of 14 740 samples from 12 of the most abundant variants of human poly(A) motifs. Compared with one of the previous state-of-the-art methods in the literature (the random forest model with expert-crafted features), our method reduces the average error rate, false-negative rate and false-positive rate by 26, 15 and 35%, respectively. Meanwhile, our method makes ~30% fewer error predictions relative to the other string kernels. Furthermore, our method can be used to visualize the importance of oligomers and positions in predicting poly(A) motifs, from which we can observe a number of characteristics in the surrounding regions of true and false motifs that have not been reported before.

AVAILABILITY

http://sfb.kaust.edu.sa/Pages/Software.aspx.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Alternative polyadenylation sites reveal distinct chromatin accessibility and histone modification in human cell lines

MOTIVATION

In addition to alternative splicing, alternative polyadenylation has also been identified as a critical and prevalent regulatory mechanism in human gene expression. However, the mechanism of alternative polyadenylation selection and the involved factors is still largely unknown.

RESULTS

We use the ENCODE data to scan DNA functional elements, including chromatin accessibility and histone modification, around transcript cleavage sites. Our results demonstrate that polyadenylation sites tend to be less sensitive to DNase I. However, these polyadenylation sites have preference in nucleosome-depleted regions, indicating the involvement of chromatin higher-order structure rather than nucleosomes in the resultant lower chromatin accessibility. More interestingly, for genes using two polyadenylation sites, the distal sites show even lower chromatin accessibility compared with the proximal sites or the unique sites of genes using only one polyadenylation site. We also observe that the histone modification mark, histone H3 lysine 36 tri-methylation (H3K36Me3), exhibits different patterns around the cleavage sites of genes using multiple polyadenylation sites from those of genes using a single polyadenylation site. Surprisingly, the H3K36Me3 levels are comparable among the alternative polyadenylation sites themselves. In summary, polyadenylation and alternative polyadenylation are closely related to functional elements on the DNA level.

CONTACT

liang.chen@usc.edu.

In silico prediction of mRNA poly(A) sites in Chlamydomonas reinhardtii

Accurately predicting polyadenylation [poly(A)] sites is important for defining the end of genes and understanding gene regulation mechanisms. Alternative polyadenylation (APA) has been demonstrated to play an important role in transcriptome diversity and regulating gene expression. To accurately predict poly(A) and APA sites in Chlamydomonas reinhardtii, a green alga that can be used to produce renewable energy, we proposed a novel model that integrated five methods for representing the features of these sites with a combined classifier. We presented a new grouping method based on pattern assembly to classify the poly(A) sites into four groups. We used five methods, involving the predicted RNA secondary structure, the term frequency-inverse document frequency weight, first-order Markov chain, pentamer ratio and a position weight matrix, to generate the feature space. We then developed a heuristic method to form the combined classifier by weighting multiple classifiers to predict poly(A) sites in each group. The high specificity and sensitivity of this model were demonstrated by testing the four groups of poly(A) sites and the intronic APA sites. The average prediction performance was approximately 8 % higher than the performance of a previous prediction model. For the group without any conserved patterns, the prediction accuracy was 9 % higher than for the accuracy with the previous technique. However, the prediction efficiency of this group was still significantly lower than that of the other groups, indicating the importance of identifying additional signal patterns for poly(A) site prediction. We also predicted the alternative poly(A) sites in introns with good accuracy. This prediction model was designed to be easily expanded with new classifiers or new features. Therefore, this model is applicable to new data or other species. Our model will be useful both in genome annotation because it predicts the end of a mature transcript and in genetic engineering because it enables researchers to eliminate undesirable poly(A) sites.