One of the most difficult speech recognition tasks is accurate recognition of human to human communication. Advances in deep learning over the last few years have produced major speech recognition improvements on the representative Switchboard conversational corpus. Word error rates that just a few years ago were 14% have dropped to 8.0%, then 6.6% and most recently 5.8%, and are now believed to be within striking range of human performance. This then raises two issues -what IS human performance, and how far down can we still drive speech recognition error rates? A recent paper by Microsoft suggests that we have already achieved human performance. In trying to verify this statement, we performed an independent set of human performance measurements on two conversational tasks and found that human performance may be considerably better than what was earlier reported, giving the community a significantly harder goal to achieve. We also report on our own efforts in this area, presenting a set of acoustic and language modeling techniques that lowered the word error rate of our own English conversational telephone LVCSR system to the level of 5.5%/10.3% on the Switchboard/CallHome subsets of the Hub5 2000 evaluation, which -at least at the writing of this paper -is a new performance milestone (albeit not at what we measure to be human performance!). On the acoustic side, we use a score fusion of three models: one LSTM with multiple feature inputs, a second LSTM trained with speaker-adversarial multitask learning and a third residual net (ResNet) with 25 convolutional layers and time-dilated convolutions. On the language modeling side, we use word and character LSTMs and convolutional WaveNet-style language models.

N 6-methyladenosine (m6A) and N6, 2′-O-Dimethyladenosine (m6Am) modifications (m6A/m) of messenger RNA mediate diverse cellular functions. Oncogenic Kaposi’s sarcoma-associated herpesvirus (KSHV) has latent and lytic replication phases that are essential for the development of KSHV-associated cancers. To date, the role of m6A/m in KSHV replication and tumorigenesis is unclear. Here, we provide mechanistic insights by examining the viral and cellular m6A/m epitranscriptomes during KSHV latent and lytic infection. KSHV transcripts contain abundant m6A/m modifications during latent and lytic replication, and these modifications are highly conserved among different cell types and infection systems. Knockdown of YTHDF2 enhanced lytic replication by impeding KSHV RNA degradation. YTHDF2 binds to viral transcripts and differentially mediates their stability. KSHV latent infection induces 5′UTR hypomethylation and 3′UTR hypermethylation of the cellular epitranscriptome, regulating oncogenic and epithelial-mesenchymal transition pathways. KSHV lytic replication induces dynamic reprograming of epitranscriptome, regulating pathways that control lytic replication. These results reveal a critical role of m6A/m modifications in KSHV lifecycle and provide rich resources for future investigations.

The MATLAB package 'exomePeak' and additional details are available at http://compgenomics.utsa.edu/exomePeak/.

Kaposi's sarcoma-associated herpesvirus (KSHV) is causally linked to several human cancers, including Kaposi's sarcoma, primary effusion lymphoma and multicentric Castleman's disease, malignancies commonly found in HIV-infected patients. While KSHV encodes diverse functional products, its mechanism of oncogenesis remains unknown. In this study, we determined the roles KSHV microRNAs (miRs) in cellular transformation and tumorigenesis using a recently developed KSHV-induced cellular transformation system of primary rat mesenchymal precursor cells. A mutant with a cluster of 10 precursor miRs (pre-miRs) deleted failed to transform primary cells, and instead, caused cell cycle arrest and apoptosis. Remarkably, the oncogenicity of the mutant virus was fully restored by genetic complementation with the miR cluster or several individual pre-miRs, which rescued cell cycle progression and inhibited apoptosis in part by redundantly targeting IκBα and the NF-κB pathway. Genomic analysis identified common targets of KSHV miRs in diverse pathways with several cancer-related pathways preferentially targeted. These works define for the first time an essential viral determinant for KSHV-induced oncogenesis and identify NF-κB as a critical pathway targeted by the viral miRs. Our results illustrate a common theme of shared functions with hierarchical order among the KSHV miRs.

Modifications of histone proteins have essential roles in normal development and human disease. Recognition of modified histones by 'reader' proteins is a key mechanism that mediates the function of histone modifications, but how the dysregulation of these readers might contribute to disease remains poorly understood. We previously identified the ENL protein as a reader of histone acetylation via its YEATS domain, linking it to the expression of cancer-driving genes in acute leukaemia 1 . Recurrent hotspot mutations have been found in the ENL YEATS domain in Wilms tumour 2,3 , the most common type of paediatric kidney cancer. Here we show, using human and mouse cells, that these mutations impair cell-fate regulation by conferring gain-of-function in chromatin recruitment and transcriptional control. ENL mutants induce gene-expression changes that favour a premalignant cell fate, and, in an assay for nephrogenesis using murine cells, result in undifferentiated structures resembling those observed in human Wilms tumour. Mechanistically, although bound to largely similar genomic loci as the wild-type protein, ENL mutants exhibit increased occupancy at a subset of targets, leading to a marked increase in the recruitment and activity of transcription elongation machinery that enforces active transcription from target loci. Furthermore, ectopically expressed ENL mutants exhibit greater self-association and form discrete and dynamic nuclear puncta that are characteristic of biomolecular hubs consisting of local high concentrations of regulatory factors. Such mutation-driven ENL self-association is functionally linked to enhanced chromatin occupancy and gene activation. Collectively, our findings show that hotspot mutations in a chromatin-reader domain drive self-reinforced recruitment, derailing normal cell-fate control during development and leading to an oncogenic outcome.The eleven-nineteen-leukaemia protein (ENL) is a chromatin reader that maintains the oncogenic state in leukaemia 1,4 . ENL interacts with acetylated histone proteins via its well conserved YEATS (Yaf9, ENL, AF9, Taf14, Sas5) domain, and, in so doing, helps to recruit and stabilize its associated transcriptional machinery to drive the transcription of target genes. Recently, somatic mutations in the ENL gene (also known as MLLT1) were found in about 5% of people with Wilms tumour, making ENL one of the most frequently mutated genes in this cancer type. These mutations are recurrent, heterozygous and highly clustered in the ENL YEATS domain. Interestingly, these 'hotspot' mutations all involve small inframe insertions or deletions (Fig. 1a and Extended Data Fig. 1a). Whether and how such ENL mutations promote the formation of Wilms tumour was unclear and is the focus of our study. Wan et al.

BackgroundCancers have long been recognized to be not only genetically but also epigenetically distinct from their tissues of origin. Although genetic alterations underlying oncogene upregulation have been well studied, to what extent epigenetic mechanisms, such as DNA methylation, can also induce oncogene expression remains unknown.ResultsHere, through pan-cancer analysis of 4174 genome-wide profiles, including whole-genome bisulfite sequencing data from 30 normal tissues and 35 solid tumors, we discover a strong correlation between gene-body hypermethylation of DNA methylation canyons, defined as broad under-methylated regions, and overexpression of approximately 43% of homeobox genes, many of which are also oncogenes. To gain insights into the cause-and-effect relationship, we use a newly developed dCas9-SunTag-DNMT3A system to methylate genomic sites of interest. The locus-specific hypermethylation of gene-body canyon, but not promoter, of homeobox oncogene DLX1, can directly increase its gene expression.ConclusionsOur pan-cancer analysis followed by functional validation reveals DNA hypermethylation as a novel epigenetic mechanism for homeobox oncogene upregulation.Electronic supplementary materialThe online version of this article (10.1186/s13059-018-1492-3) contains supplementary material, which is available to authorized users.

N6-Methyladenosine (mA) transcriptome methylation is an exciting new research area that just captures the attention of research community. We present in this paper, MeTDiff, a novel computational tool for predicting differential mA methylation sites from Methylated RNA immunoprecipitation sequencing (MeRIP-Seq) data. Compared with the existing algorithm exomePeak, the advantages of MeTDiff are that it explicitly models the reads variation in data and also devices a more power likelihood ratio test for differential methylation site prediction. Comprehensive evaluation of MeTDiff's performance using both simulated and real datasets showed that MeTDiff is much more robust and achieved much higher sensitivity and specificity over exomePeak. The R package "MeTDiff" and additional details are available at: https://github.com/compgenomics/MeTDiff.

Biological features, such as genes and transcription factor binding sites, are often denoted with genome-based coordinates as the genomic features. While genome-based representation is usually very effective in correlating various biological features, it can be tedious to examine the relationship between RNA-related genomic features and the landmarks of RNA transcripts with existing tools due to the difficulty in the conversion between genome-based coordinates and RNA-based coordinates. We developed here an open source Guitar R/Bioconductor package for sketching the transcriptomic view of RNA-related biological features represented by genome based coordinates. Internally, Guitar package extracts the standardized RNA coordinates with respect to the landmarks of RNA transcripts, with which hundreds of millions of RNA-related genomic features can then be efficiently analyzed within minutes. We demonstrated the usage of Guitar package in analyzing posttranscriptional RNA modifications (5-methylcytosine and N6-methyladenosine) derived from high-throughput sequencing approaches (MeRIP-Seq and RNA BS-Seq) and show that RNA 5-methylcytosine (m5C) is enriched in 5′UTR. The newly developed Guitar R/Bioconductor package achieves stable performance on the data tested and revealed novel biological insights. It will effectively facilitate the analysis of RNA methylation data and other RNA-related biological features in the future.