Deep learning of the tissue-regulated splicing code

Leung, M.; Xiong, Hui; Lee, Leo J.; Frey, Brendan J.

doi:10.1093/bioinformatics/btu277

Cited by 415 publications

(268 citation statements)

References 20 publications

Supporting

Mentioning

260

Contrasting

Unclassified

Order By: Relevance

“…Machine learning methods use these and other genomic features trained against genome-wide RNA sequencing datasets to build predictive models of splicing regulation [7][8][9] . However, the predictive power of these models may come almost entirely from sequence conservation rather than the mechanistic understanding of splicing 18,19 . These models predict that human genetic variation, and especially rare variation, often disrupt sequence features required for proper exon recognition, but it is difficult to verify the accuracy of these predictions at large scales Several groups have developed massively parallel reporter assays of splicing 8,14,20,21 .…”

Section: Main Textmentioning

confidence: 99%

Many rare genetic variants have unrecognized large-effect disruptions to exon recognition

Cheung

Yao

et al. 2017

Preprint

View full text Add to dashboard Cite

Any individual’s genome contains ∼4-5 million genetic variants that differ from reference, and understanding how these variants give rise to trait diversity and disease susceptibility is a central goal of human genetics1. A vast majority (96-99%) of an individual’s variants are common, though at a population level the overwhelming majority of variants are rare2–5. Because of their scarcity in an individual’s genome, rare variants that play important roles in complex traits are likely to have large functional effects6,7. Mutations that cause an exon to be skipped can have severe functional consequences on gene function, and many known disease-causing mutations reduce or eliminate exon recognition8. Here we explore the extent to which rare genetic variation in humans results in near complete loss of exon recognition. We developed a Multiplexed Functional Assay of Splicing using Sort-seq (MFASS) that allows us to measure exon inclusion in thousands of human exons and surrounding intronic sequence simultaneously. We assayed 27,733 extant variants in the Exome Aggregation Consortium (ExAC)9 within or adjacent to 2,339 human exons, and found that 3.8% (1,050) of the variants, almost all of which were extremely rare, led to large-effect defects in exon recognition. Importantly, we find that 83% of these splice-disrupting variants (SDVs) are located outside of canonical splice sites, are distributed evenly across distinct exonic and intronic regions, and are difficult to predict a priori. Our results indicate that loss of exon recognition is an important and underappreciated means by which rare variants exert large functional effects, and that MFASS enables their empirical assessment for splicing defects at scale.

show abstract

Section: Main Textmentioning

confidence: 99%

Many rare genetic variants have unrecognized large-effect disruptions to exon recognition

Cheung

Yao

et al. 2017

Preprint

View full text Add to dashboard Cite

show abstract

“…The deep learning (DL) method has emerged as the state-of-the-art technique for genomic sequence analysis [64]. A deep neural network (DNN) is one of implementations in DL, which generally refers to methods that map data through multiple levels of a feed-forward neural network to reveal some intractable and non-linear relation between input data and hidden factors.…”

Section: Traditional and Deep Neural Networkmentioning

confidence: 99%

“…However, mature concepts about deep learning, including a deep neural network, were not proposed until the mid-2000s [66][67][68]. Since then, only a small number of works [65,69,70] have applied deep learning in the life sciences, even though it has shown tremendous promise [64]. As a promising method for the future directions of gene prediction, deep learning and other emerging methods are further discussed in Section 7.…”

Section: Traditional and Deep Neural Networkmentioning

confidence: 99%

A Comprehensive Review of Emerging Computational Methods for Gene Identification

2016

J Inf Process Syst

View full text Add to dashboard Cite

Gene identification is at the center of genomic studies. Although the first phase of the Encyclopedia of DNA Elements (ENCODE) project has been claimed to be complete, the annotation of the functional elements is far from being so. Computational methods in gene identification continue to play important roles in this area and other relevant issues. So far, a lot of work has been performed on this area, and a plethora of computational methods and avenues have been developed. Many review papers have summarized these methods and other related work. However, most of them focus on the methodologies from a particular aspect or perspective. Different from these existing bodies of research, this paper aims to comprehensively summarize the mainstream computational methods in gene identification and tries to provide a short but concise technical reference for future studies. Moreover, this review sheds light on the emerging trends and cutting-edge techniques that are believed to be capable of leading the research on this field in the future.

show abstract

“…Indeed, solving image classification problems is not the only way deep learning can assist biomedical researches. As some recent works show [12,18], deep neural networks are being used for predictive modeling, using RNA-Seq data as input. We may ask ourselves, though, whether the use of deep learning, when it comes to produce predictive models, is as straight forward as it is in other kind of problems.…”

Section: Introductionmentioning

confidence: 99%

Deep Learning to Analyze RNA-Seq Gene Expression Data

Urda

Montes-Torres

Moreno

et al. 2017

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. Deep learning models are currently being applied in several areas with great success. However, their application for the analysis of high-throughput sequencing data remains a challenge for the research community due to the fact that this family of models are known to work very well in big datasets with lots of samples available, just the opposite scenario typically found in biomedical areas. In this work, a first approximation on the use of deep learning for the analysis of RNA-Seq gene expression profiles data is provided. Three public cancer-related databases are analyzed using a regularized linear model (standard LASSO) as baseline model, and two deep learning models that differ on the feature selection technique used prior to the application of a deep neural net model. The results indicate that a straightforward application of deep nets implementations available in public scientific tools and under the conditions described within this work is not enough to outperform simpler models like LASSO. Therefore, smarter and more complex ways that incorporate prior biological knowledge into the estimation procedure of deep learning models may be necessary in order to obtain better results in terms of predictive performance.

show abstract

Deep learning of the tissue-regulated splicing code

Cited by 415 publications

References 20 publications

Many rare genetic variants have unrecognized large-effect disruptions to exon recognition

Many rare genetic variants have unrecognized large-effect disruptions to exon recognition

A Comprehensive Review of Emerging Computational Methods for Gene Identification

Deep Learning to Analyze RNA-Seq Gene Expression Data

Contact Info

Product

Resources

About