Deep learning for DNase I hypersensitive sites identification

Background

The DNase I hypersensitive sites (DHSs) are associated with the cis-regulatory DNA elements. An efficient method of identifying DHSs can enhance the understanding on the accessibility of chromatin. Despite a multitude of resources available on line including experimental datasets and computational tools, the complex language of DHSs remains incompletely understood.

Methods

Here, we address this challenge using an approach based on a state-of-the-art machine learning method. We present a novel convolutional neural network (CNN) which combined Inception like networks with a gating mechanism for the response of multiple patterns and longterm association in DNA sequences to predict multi-scale DHSs in Arabidopsis, rice and Homo sapiens.

Results

Our method obtains 0.961 area under curve (AUC) on Arabidopsis, 0.969 AUC on rice and 0.918 AUC on Homo sapiens.

Conclusions

Our method provides an efficient and accurate way to identify multi-scale DHSs sequences by deep learning.

     

In vivo analysis of DNase I hypersensitive sites in the human CFTR gene.

BACKGROUND: The cystic fibrosis transmembrane conductance regulator gene (CFTR) shows a complex pattern of expression. The regulatory elements conferring tissue-specific and temporal regulation are thought to lie mainly outside the promoter region. Previously, we identified DNase I hypersensitive sites (DHS) that may contain regulatory elements associated with the CFTR gene at -79.5 and at -20.5 kb with respect to the ATG and at 10 kb into the first intron. MATERIALS AND METHODS: In order to evaluate these regulatory elements in vivo we examined these DHS in a human CFTR gene that was introduced on a yeast artificial chromosome (YAC) into transgenic mice. The 310 kb human CFTR YAC was shown to restore the pheno-type of CF-null mice and so is likely to contain most of the regulatory elements required for tissue-specific expression of CFTR. RESULTS: We found that the YAC does not include the -79.5 kb region. The DHS at -20.5 kb is present in the chromatin of most tissues of the transgenic mice, supporting its non-tissue-specific nature. The DHS in the first intron is present in a more restricted set of tissues in the mice, although its presence does not show complete concordance with CFTR expression. The intron I DHS may be important for the higher levels of expression found in human pancreatic ducts and in lung submucosal glands. CONCLUSION: These data support the in vivo importance of these regulatory elements.     

DNase I hypersensitive sites and transcriptional activation of the lamin A/C gene.

The lamin A/C gene encodes subtypes of nuclear lamins, which are involved in nuclear envelope formation, and was recently identified as the responsible gene for the autosomal dominant Emery-Dreifuss muscular dystrophy. Expression of the lamin A/C gene is developmentally regulated but little is known about the regulatory mechanism. Previous studies of lamin A/C expression suggested that the chromatin structure is important for the regulation of its expression. To elucidate the regulatory mechanism of the lamin A/C gene expression, we have analysed the functional region of the mouse lamin A/C promoter and the chromatin structure of the gene in terms of nucleosome structure and DNase I hypersensitivity. Our analyses revealed disruption of the nucleosome array at the promoter region and the presence of multiple DNase I hypersensitive sites (HSs) which were specifically associated with expression of the lamin A/C gene. Inclusion of a segment which contained the HSs in a lamin A/C promoter-luciferase reporter plasmid showed no effect on the transfected promoter activity in transient expression assays. On the other hand, substantial enhancement of the promoter activity was detected when the transfected DNA was stably integrated into the genome, suggesting the importance of the HSs in the regulation of lamin A/C expression.     

Condensation of chromatin into chromosomes preserves an open configuration but alters the DNase I hypersensitive cleavage sites of the transcribed gene

DNase I was used to probe the molecular organization of the chicken ovalbumin (OV) gene and glyceraldehyde 3-phosphate dehydrogenase (GPD) gene in interphase nuclei and in metaphase chromosomes of cultured chicken lymphoblastoid cells (MSB-1 line). The OV gene was not transcribed in this cell line, whereas the GPD gene was constitutively expressed. The GPD gene was more sensitive to DNase I digestion than the OV gene in both interphase nuclei and metaphase chromosomes, as determined by Southern blotting and liquid hybridization techniques. In addition, we observed DNase I hypersensitive sites around the 5' region of the GPD gene. These hypersensitive sites were not always at the same locations between the interphase nuclei and metaphase chromosomes. Our results suggest that chromatin condensation and decondensation during cell cycle alters nuclease hypersensitive cleavage sites.     

Identification of insulin-induced sites of ribosomal protein S6 phosphorylation in Drosophila melanogaster

Insulin treatment of Drosophila melanogaster Kc 167 cells induces the multiple phosphorylation of a Drosophila ribosomal protein, as judged by its decreased electrophoretic mobility on two-dimensional polyacrylamide gels. The extent to which insulin induces this response is potentiated by cycloheximide and blocked by pretreatment with rapamycin. Isolation and mass spectrometric analysis revealed that the multiply phosphorylated protein was the larger of two Drosophila melanogaster orthologues of mammalian 40S ribosomal protein S6, termed here DS6A. Proteolytic cleavage of DS6A derived from stimulated Kc 167 cells with the endoproteinase Lys-C released a number of peptides, one of which contained all the putative phosphorylation sites. Conversion of phosphoserines to dehydroalanines with Ba(OH)(2) showed that the sites of phosphorylation reside at the carboxy terminus of DS6A. The sites of phosphorylation were identified by Edman degradation after conversion of the phosphoserine residues to S-ethylcysteine as Ser(233), Ser(235), Ser(239), Ser(242), and Ser(245). Finally, phosphopeptide mapping of individual phosphoderivatives, isolated from two-dimensional polyacrylamide gels, indicated that DS6A phosphorylation, in analogy to mammalian S6 phosphorylation, appears to proceed in an ordered fashion. The importance of these observations in cell growth and development is discussed.     

Sequence, structure, and codon preference of the Drosophila ribosomal protein 49 gene.

In this communication, we describe several features of the D. melanogaster gene which codes for ribosomal protein 49 (rp49). Nucleotide sequence analysis in conjunction with primer extension and S1 nuclease protection experiments show that the structure of the rp49 gene consists of a 102 bp 5' exon, a single 59 bp intron, and a 420 bp 3' exon, encoding a total of 132 amino acids. The rp49 gene shares many features with other abundantly expressed Drosophila genes, including codon preference, which are discussed.     

The Drosophila melanogaster RPS17 gene encoding ribosomal protein S17

A human ribosomal protein S17 cDNA [Chen et al., Proc. Natl. Acad. Sci. USA 83 (1986) 6907-6911] was used as heterologous probe to isolate S17 clones from Drosophila genomic and cDNA recombinant libraries. Five S17 genomic clones were recognized; all contained overlapping regions of a single chromosomal site. Subsequently the Drosophila RPS17 gene was mapped by in situ hybridization to chromosome 3L, band 67B1-5. The locus spans approximately 1000 bp of DNA and includes four exons. It is preceded by conventional CAAT and TATA RNA polymerase II promoter motifs. The 131 amino acid protein encoded within Drosophila RPS17 is similar to ribosomal proteins from several other eukaryotes. Comparison of eukaryotic S17 proteins' primary structures as well as the number and location of their genes' intervening sequences suggest that S17 is a relatively recent addition to the ribosomal protein family, probably post-dating divergence of eukaryotes and prokaryotes.     

Cardiomyopathy is associated with ribosomal protein gene haplo-insufficiency in Drosophila melanogaster

The Minute syndrome in Drosophila melanogaster is characterized by delayed development, poor fertility, and short slender bristles. Many Minute loci correspond to disruptions of genes for cytoplasmic ribosomal proteins, and therefore the phenotype has been attributed to alterations in translational processes. Although protein translation is crucial for all cells in an organism, it is unclear why Minute mutations cause effects in specific tissues. To determine whether the heart is sensitive to haplo-insufficiency of genes encoding ribosomal proteins, we measured heart function of Minute mutants using optical coherence tomography. We found that cardiomyopathy is associated with the Minute syndrome caused by haplo-insufficiency of genes encoding cytoplasmic ribosomal proteins. While mutations of genes encoding non-Minute cytoplasmic ribosomal proteins are homozygous lethal, heterozygous deficiencies spanning these non-Minute genes did not cause a change in cardiac function. Deficiencies of genes for non-Minute mitochondrial ribosomal proteins also did not show abnormal cardiac function, with the exception of a heterozygous disruption of mRpS33. We demonstrate that cardiomyopathy is a common trait of the Minute syndrome caused by haplo-insufficiency of genes encoding cytoplasmic ribosomal proteins. In contrast, most cases of heterozygous deficiencies of genes encoding non-Minute ribosomal proteins have normal heart function in adult Drosophila.