Deep learning describes a class of machine learning algorithms that are capable of combining raw inputs into layers of intermediate features. These algorithms have recently shown impressive results across a variety of domains. Biology and medicine are data-rich disciplines, but the data are complex and often ill-understood. Hence, deep learning techniques may be particularly well suited to solve problems of these fields. We examine applications of deep learning to a variety of biomedical problems-patient classification, fundamental biological processes and treatment of patients-and discuss whether deep learning will be able to transform these tasks or if the biomedical sphere poses unique challenges. Following from an extensive literature review, we find that deep learning has yet to revolutionize biomedicine or definitively resolve any of the most pressing challenges in the field, but promising advances have been made on the prior state of the art. Even though improvements over previous baselines have been modest in general, the recent progress indicates that deep learning methods will provide valuable means for speeding up or aiding human investigation. Though progress has been made linking a specific neural network's prediction to input features, understanding how users should interpret these models to make testable hypotheses about the system under study remains an open challenge. Furthermore, the limited amount of labelled data for training presents problems in some domains, as do legal and privacy constraints on work with sensitive health records. Nonetheless, we foresee deep learning enabling changes at both bench and bedside with the potential to transform several areas of biology and medicine.

https://royalsocietypublishing.org/doi/pdf/10.1098/rsif.2017.0387

Opportunities and obstacles for deep learning in biology and medicine.

Over the past decade, RNA sequencing (RNA-seq) has become an indispensable tool for transcriptome-wide analysis of differential gene expression and differential splicing of mRNAs. However, as next-generation sequencing technologies have developed, so too has RNA-seq. Now, RNA-seq methods are available for studying many different aspects of RNA biology, including single-cell gene expression, translation (the translatome) and RNA structure (the structurome). Exciting new applications are being explored, such as spatial transcriptomics (spatialomics). Together with new long-read and direct RNA-seq technologies and better computational tools for data analysis, innovations in RNA-seq are contributing to a fuller understanding of RNA biology, from questions such as when and where transcription occurs to the folding and intermolecular interactions that govern RNA function.

RNA sequencing: the teenage years

We developed PolyA-seq, a strand-specific and quantitative method for high-throughput sequencing of 3′ ends of polyadenylated transcripts, and used it to globally map polyadenylation (polyA) sites in 24 matched tissues in human, rhesus, dog, mouse, and rat. We show that PolyA-seq is as accurate as existing RNA sequencing (RNA-seq) approaches for digital gene expression (DGE), enabling simultaneous mapping of polyA sites and quantitative measurement of their usage. In human, we confirmed 158,533 known sites and discovered 280,857 novel sites (FDR < 2.5%). On average 10% of novel human sites were also detected in matched tissues in other species. Most novel sites represent uncharacterized alternative polyA events and extensions of known transcripts in human and mouse, but primarily delineate novel transcripts in the other three species. A total of 69.1% of known human genes that we detected have multiple polyA sites in their 3′UTRs, with 49.3% having three or more. We also detected polyadenylation of noncoding and antisense transcripts, including constitutive and tissue-specific primary microRNAs. The canonical polyA signal was strongly enriched and positionally conserved in all species. In general, usage of polyA sites is more similar within the same tissues across different species than within a species. These quantitative maps of polyA usage in evolutionarily and functionally related samples constitute a resource for understanding the regulatory mechanisms underlying alternative polyadenylation.

/pdf/a-quantitative-atlas-of-polyadenylation-in-five-mammals-4415scdp8e.pdf

A quantitative atlas of polyadenylation in five mammals

http://medinalab.org/new/wp-content/uploads/2013/10/Symbiodinium_minutum_genome_CurrentBiology2013.pdf

Draft Assembly of the Symbiodinium minutum Nuclear Genome Reveals Dinoflagellate Gene Structure

We have integrated and analyzed a large number of data sets from a variety of genomic assays using a novel computational pipeline to provide a global view of estrogen receptor 1 (ESR1; a.k.a. ERα) enhancers in MCF-7 human breast cancer cells. Using this approach, we have defined a class of primary transcripts (eRNAs) that are transcribed uni- or bidirectionally from estrogen receptor binding sites (ERBSs) with an average transcription unit length of ∼3-5 kb. The majority are up-regulated by short treatments with estradiol (i.e., 10, 25, or 40 min) with kinetics that precede or match the induction of the target genes. The production of eRNAs at ERBSs is strongly correlated with the enrichment of a number of genomic features that are associated with enhancers (e.g., H3K4me1, H3K27ac, EP300/CREBBP, RNA polymerase II, open chromatin architecture), as well as enhancer looping to target gene promoters. In the absence of eRNA production, strong enrichment of these features is not observed, even though ESR1 binding is evident. We find that flavopiridol, a CDK9 inhibitor that blocks transcription elongation, inhibits eRNA production but does not affect other molecular indicators of enhancer activity, suggesting that eRNA production occurs after the assembly of active enhancers. Finally, we show that an enhancer transcription "signature" based on GRO-seq data can be used for de novo enhancer prediction across cell types. Together, our studies shed new light on the activity of ESR1 at its enhancer sites and provide new insights about enhancer function.

/pdf/enhancer-transcripts-mark-active-estrogen-receptor-binding-31uyywceeu.pdf

Enhancer transcripts mark active estrogen receptor binding sites

We performed a genome-wide analysis of transcriptional start sites (TSSs) in human genes by multifaceted use of a massively parallel sequencer. By analyzing 800 million sequences that were obtained from various types of transcriptome analyses, we characterized 140 million TSS tags in 12 human cell types. Despite the large number of TSS clusters (TSCs), the number of TSCs was observed to decrease sharply with increasing expression levels. Highly expressed TSCs exhibited several characteristic features: Nucleosome-seq analysis revealed highly ordered nucleosome structures, ChIP-seq analysis detected clear RNA polymerase II binding signals in their surrounding regions, evaluations of previously sequenced and newly shotgun-sequenced complete cDNA sequences showed that they encode preferable transcripts for protein translation, and RNA-seq analysis of polysome-incorporated RNAs yielded direct evidence that those transcripts are actually translated into proteins. We also demonstrate that integrative interpretation of transcriptome data is essential for the selection of putative alternative promoter TSCs, two of which also have protein consequences. Furthermore, discriminative chromatin features that separate TSCs at different expression levels were found for both genic TSCs and intergenic TSCs. The collected integrative information should provide a useful basis for future biological characterization of TSCs.

/pdf/genome-wide-characterization-of-transcriptional-start-sites-2q4wawde79.pdf

Genome-wide characterization of transcriptional start sites in humans by integrative transcriptome analysis

On the basis of integrated transcriptome analysis, we show that not all transcriptional start site clusters (TSCs) in the intergenic regions (iTSCs) have the same properties; thus, it is possible to discriminate the iTSCs that are likely to have biological relevance from the other noise-level iTSCs. We used a total of 251 933 381 short-read sequence tags generated from various types of transcriptome analyses in order to characterize 6039 iTSCs, which have significant expression levels. We analyzed and found that 23% of these iTSCs were located in the proximal regions of the RefSeq genes. These RefSeq-linked iTSCs showed similar expression patterns with the neighboring RefSeq genes, had widely fluctuating transcription start sites and lacked ordered nucleosome positioning. These iTSCs seemed not to form independent transcriptional units, simply representing the by-products of the neighboring RefSeq genes, in spite of their significant expression levels. Similar features were also observed for the TSCs located in the antisense regions of the RefSeq genes. Furthermore, for the remaining iTSCs that were not associated with any RefSeq genes, we demonstrate that integrative interpretation of the transcriptome data provides essential information to specify their biological functions in the hypoxic responses of the cells.

/pdf/characterization-of-transcription-start-sites-of-putative-3uxh16iclc.pdf

Nuankanya P. Sathira

Papers

Genome-wide characterization of transcriptional start sites in humans by integrative transcriptome analysis

Characterization of Transcription Start Sites of Putative Non-coding RNAs by Multifaceted Use of Massively Paralleled Sequencer