Recent advances in convolutional neural networks

The PASCAL Visual Object Classes Challenge

http://cvssp-data.eps.surrey.ac.uk/Personal/AdrianHilton/Publications_files/survey_techrep.pdf

A survey of advances in vision-based human motion capture and analysis

In the last few years, deep learning has led to very good performance on a variety of problems, such as visual recognition, speech recognition and natural language processing. Among different types of deep neural networks, convolutional neural networks have been most extensively studied. Leveraging on the rapid growth in the amount of the annotated data and the great improvements in the strengths of graphics processor units, the research on convolutional neural networks has been emerged swiftly and achieved state-of-the-art results on various tasks. In this paper, we provide a broad survey of the recent advances in convolutional neural networks. We detailize the improvements of CNN on different aspects, including layer design, activation function, loss function, regularization, optimization and fast computation. Besides, we also introduce various applications of convolutional neural networks in computer vision, speech and natural language processing.

Recent Advances in Convolutional Neural Networks

https://www.cell.com/cell/pdf/S0092-8674(09)00148-2.pdf

RNA and Disease

We have collected over half a million splice sites from five species-Homo sapiens, Mus musculus, Drosophila melanogaster, Caenorhabditis elegans and Arabidopsis thaliana-and classified them into four subtypes: U2-type GT-AG and GC-AG and U12-type GT-AG and AT-AC. We have also found new examples of rare splice-site categories, such as U12-type introns without canonical borders, and U2-dependent AT-AC introns. The splice-site sequences and several tools to explore them are available on a public website (SpliceRack). For the U12-type introns, we find several features conserved across species, as well as a clustering of these introns on genes. Using the information content of the splice-site motifs, and the phylogenetic distance between them, we identify: (i) a higher degree of conservation in the exonic portion of the U2-type splice sites in more complex organisms; (ii) conservation of exonic nucleotides for U12-type splice sites; (iii) divergent evolution of C.elegans 3' splice sites (3'ss) and (iv) distinct evolutionary histories of 5' and 3'ss. Our study proves that the identification of broad patterns in naturally-occurring splice sites, through the analysis of genomic datasets, provides mechanistic and evolutionary insights into pre-mRNA splicing.

https://dr.ntu.edu.sg/bitstream/10220/8868/1/9.%20Comprehensive%20splice-site%20analysis%20using%20comparative%20genomics.pdf

Comprehensive splice-site analysis using comparative genomics

Mice lacking desmin produce muscle fibers with Z disks and normal sarcomeric organization. However, the muscles are mechanically fragile and degenerate upon repeated contractions. We report here a human patient with severe generalized myopathy and aberrant intrasarcoplasmic accumulation of desmin intermediate filaments. Muscle tissue from this patient lacks the wild-type desmin allele and has a desmin gene mutation encoding a 7-aa deletion within the coiled-coil segment of the protein. We show that recombinant desmin harboring this deletion cannot form proper desmin intermediate filament networks in cultured cells, nor is it able to assemble into 10-nm filaments in vitro. These findings provide direct evidence that a mutation in desmin can cause human myopathies.

/pdf/a-dysfunctional-desmin-mutation-in-a-patient-with-severe-1930zhcchj.pdf

A dysfunctional desmin mutation in a patient with severe generalized myopathy

Splice site selection is fundamental to pre-mRNA splicing and the expansion of genomic coding potential. 5' Splice sites (5'ss) are the critical elements at the 5' end of introns and are extremely diverse, as thousands of different sequences act as bona fide 5'ss in the human transcriptome. Most 5'ss are recognized by base-pairing with the 5' end of the U1 small nuclear RNA (snRNA). Here we review the history of research on 5'ss selection, highlighting the difficulties of establishing how base-pairing strength determines splicing outcomes. We also discuss recent work demonstrating that U1 snRNA:5'ss helices can accommodate noncanonical registers such as bulged duplexes. In addition, we describe the mechanisms by which other snRNAs, regulatory proteins, splicing enhancers, and the relative positions of alternative 5'ss contribute to selection. Moreover, we discuss mechanisms by which the recognition of numerous candidate 5'ss might lead to selection of a single 5'ss and propose that protein complexes propagate along the exon, thereby changing its physical behavior so as to affect 5'ss selection.

/pdf/pick-one-but-be-quick-5-splice-sites-and-the-problems-of-too-53tljo06ig.pdf

Pick one, but be quick: 5′ splice sites and the problems of too many choices

Cryptic splice sites are used only when use of a natural splice site is disrupted by mutation. To determine the features that distinguish authentic from cryptic 5' splice sites (5'ss), we systematically analyzed a set of 76 cryptic 5'ss derived from 46 human genes. These cryptic 5'ss have a similar frequency distribution in exons and introns, and are usually located close to the authentic 5'ss. Statistical analysis of the strengths of the 5'ss using the Shapiro and Senapathy matrix revealed that authentic 5'ss have significantly higher score values than cryptic 5'ss, which in turn have higher values than the mutant ones. beta-Globin provides an interesting exception to this rule, so we chose it for detailed experimental analysis in vitro. We found that the sequences of the beta-globin authentic and cryptic 5'ss, but not their surrounding context, determine the correct 5'ss choice, although their respective scores do not reflect this functional difference. Our analysis provides a statistical basis to explain the competitive advantage of authentic over cryptic 5'ss in most cases, and should facilitate the development of tools to reliably predict the effect of disease-associated 5'ss-disrupting mutations at the mRNA level.

/pdf/intrinsic-differences-between-authentic-and-cryptic-5-splice-2l0gsl6v1r.pdf

Intrinsic differences between authentic and cryptic 5′ splice sites

We previously showed that the authentic 5' splice site (5'ss) of the first exon in the human beta-globin gene is intrinsically stronger than a cryptic 5'ss located 16 nucleotides upstream. Here we examined by mutational analysis the contribution of individual 5'ss nucleotides to discrimination between these two 5'ss. Based on the in vitro splicing efficiencies of a panel of 26 wild-type and mutant substrates in two separate 5'ss competition assays, we established a hierarchy of 5'ss and grouped them into three functional subclasses: strong, intermediate, and weak. Competition between two 5'ss from different subclasses always resulted in selection of the 5'ss that belongs to the stronger subclass. Moreover, each subclass has different characteristic features. Strong and intermediate 5'ss can be distinguished by their predicted free energy of base-pairing to the U1 snRNA 5' terminus (DeltaG). Whereas the extent of splicing via the strong 5'ss correlates well with the DeltaG, this is not the case for competition between intermediate 5'ss. Weak 5'ss were used only when the competing authentic 5'ss was inactivated by mutation. These results indicate that extensive complementarity to U1 snRNA exerts a dominant effect for 5'ss selection, but in the case of competing 5'ss with similarly modest complementarity to U1, the role of other 5'ss features is more prominent. This study reveals the importance of additional submotifs present in certain 5'ss sequences, whose characterization will be critical for understanding 5'ss selection in human genes.

/pdf/determinants-of-the-inherent-strength-of-human-5-splice-2zos14yath.pdf

Xavier Roca

Papers

Comprehensive splice-site analysis using comparative genomics

A dysfunctional desmin mutation in a patient with severe generalized myopathy

Pick one, but be quick: 5′ splice sites and the problems of too many choices

Intrinsic differences between authentic and cryptic 5′ splice sites

Determinants of the inherent strength of human 5′ splice sites