Artificial intelligence at the national eye institute.

PURPOSE OF REVIEW: This review highlights the artificial intelligence, machine learning, and deep learning initiatives supported by the National Institutes of Health (NIH) and the National Eye Institute (NEI) and calls attention to activities and goals defined in the NEI Strategic Plan as well as opportunities for future activities and breakthroughs in ophthalmology. RECENT FINDINGS: Ophthalmology is at the forefront of artificial intelligence-based innovations in biomedical research that may lead to improvement in early detection and surveillance of ocular disease, prediction of progression, and improved quality of life. Technological advances have ushered in an era where unprecedented amounts of information can be linked that enable scientific discovery. However, there remains an unmet need to collect, harmonize, and share data in a machine actionable manner. Similarly, there is a need to ensure that efforts promote health and research equity by expanding diversity in the data and workforce. SUMMARY: The NIH/NEI has supported the development artificial intelligence-based innovations to advance biomedical research. The NIH/NEI has defined activities to achieve these goals in the NIH Strategic Plan for Data Science and the NEI Strategic Plan and have spearheaded initiatives to facilitate research in these areas.

Visual and linguistic semantic representations are aligned at the border of human visual cortex.

Semantic information in the human brain is organized into multiple networks, but the fine-grain relationships between them are poorly understood. In this study, we compared semantic maps obtained from two functional magnetic resonance imaging experiments in the same participants: one that used silent movies as stimuli and another that used narrative stories. Movies evoked activity from a network of modality-specific, semantically selective areas in visual cortex. Stories evoked activity from another network of semantically selective areas immediately anterior to visual cortex. Remarkably, the pattern of semantic selectivity in these two distinct networks corresponded along the boundary of visual cortex: for visual categories represented posterior to the boundary, the same categories were represented linguistically on the anterior side. These results suggest that these two networks are smoothly joined to form one contiguous map.

Hookah Smoke Mediates Cancer-Associated Epigenomic and Transcriptomic Signatures in Human Respiratory Epithelial Cells.

INTRODUCTION: Although communal smoking of hookah by means of water pipes is perceived to be a safe alternative to cigarette smoking, the effects of hookah smoke in respiratory epithelia have not been well characterized. This study evaluated epigenomic and transcriptomic effects of hookah smoke relative to cigarette smoke in human respiratory epithelial cells. METHODS: Primary normal human small airway epithelial cells from three donors and cdk4 and hTERT-immortalized small airway epithelial cells and human bronchial epithelial cells were cultured for 5 days in normal media with or without cigarette smoke condensates (CSCs) or water pipe condensates (WPCs). Cell count, immunoblot, RNA sequencing, quantitative real-time reverse-transcriptase polymerase chain reaction, methylation-specific polymerase chain reaction, and quantitative chromatin immunoprecipitation techniques were used to compare effects of hookah and cigarette smoke on cell proliferation, global histone marks, gene expression, and promoter-related chromatin structure. RESULTS: CSC and WPC decreased global H4K16ac and H4K20me3 histone marks and mediated distinct and overlapping cancer-associated transcriptome signatures and pathway modulations that were cell line dependent and stratified across lung cancer cells in a histology-specific manner. Epiregulin encoding a master regulator of EGFR signaling that is overexpressed in lung cancers was up-regulated, whereas FILIP1L and ABI3BP encoding mediators of senescence that are repressed in lung cancers were down-regulated by CSC and WPC. Induction of epiregulin and repression of FILIP1L and ABI3BP by these condensates coincided with unique epigenetic alterations within the respective promoters. CONCLUSIONS: These findings support translational studies to ascertain if hookah-mediated epigenomic and transcriptomic alterations in cultured respiratory epithelia are detectable and clinically relevant in hookah smokers.

Voxel-Based State Space Modeling Recovers Task-Related Cognitive States in Naturalistic fMRI Experiments.

Complex natural tasks likely recruit many different functional brain networks, but it is difficult to predict how such tasks will be represented across cortical areas and networks. Previous electrophysiology studies suggest that task variables are represented in a low-dimensional subspace within the activity space of neural populations. Here we develop a voxel-based state space modeling method for recovering task-related state spaces from human fMRI data. We apply this method to data acquired in a controlled visual attention task and a video game task. We find that each task induces distinct brain states that can be embedded in a low-dimensional state space that reflects task parameters, and that attention increases state separation in the task-related subspace. Our results demonstrate that the state space framework offers a powerful approach for modeling human brain activity elicited by complex natural tasks.

High-speed switchable lens enables the development of a volumetric stereoscopic display.

Stereoscopic displays present different images to the two eyes and thereby create a compelling three-dimensional (3D) sensation. They are being developed for numerous applications including cinema, television, virtual prototyping, and medical imaging. However, stereoscopic displays cause perceptual distortions, performance decrements, and visual fatigue. These problems occur because some of the presented depth cues (i.e., perspective and binocular disparity) specify the intended 3D scene while focus cues (blur and accommodation) specify the fixed distance of the display itself. We have developed a stereoscopic display that circumvents these problems. It consists of a fast switchable lens synchronized to the display such that focus cues are nearly correct. The system has great potential for both basic vision research and display applications.

Expressed sequence tag analysis of guinea pig (Cavia porcellus) eye tissues for NEIBank.

PURPOSE: To characterize gene expression patterns in guinea pig ocular tissues and identify orthologs of human genes from NEIBank expressed sequence tags. METHODS: RNA was extracted from dissected eye tissues of 2.5-month-old guinea pigs to make three unamplified and unnormalized cDNA libraries in the pCMVSport-6 vector for the lens, retina, and eye minus lens and retina. Over 4,000 clones were sequenced from each library and were analyzed using GRIST for clustering and gene identification. Lens crystallin EST data were validated using two-dimensional electrophoresis (2-DE), matrix assisted laser desorption (MALDI), and electrospray ionization mass spectrometry (ESIMS). RESULTS: Combined data from the three libraries generated a total of 6,694 distinctive gene clusters, with each library having between 1,000 and 3,000 clusters. Approximately 60% of the total gene clusters were novel cDNA sequences and had significant homologies to other mammalian sequences in GenBank. Complete cDNA sequences were obtained for many guinea pig lens proteins, including alphaA/alphaAinsert-, gammaN-, and gammaS-crystallins, lengsin and GRIFIN. The ratio of alphaA- to alphaB-crystallin on 2-DE gels was 8: 1 in the lens nucleus and 6.5: 1 in the cortex. Analysis of ESTs, genome sequence, and proteins (by MALDI), did not reveal any evidence for the presence of gammaD-, gammaE-, and gammaF-crystallin in the guinea pig. Predicted masses of many guinea pig lens crystallins were confirmed by ESIMS analysis. For the retina, orthologs of human phototransduction genes were found, such as Rhodopsin, S-antigen (Sag, Arrestin), and Transducin. The guinea-pig ortholog of NRL, a key rod photoreceptor-specific transcription factor, was also represented in EST data. In the 'rest-of-eye' library, the most abundant transcripts included decorin and keratin 12, representative of the cornea. CONCLUSIONS: Genomic analysis of guinea pig eye tissues provides sequence-verified clones for future studies. Guinea pig orthologs of many human eye specific genes were identified. Guinea pig gene structures were similar to their human and rodent gene counterparts. Surprisingly, no orthologs of gammaD-, gammaE-, and gammaF-crystallin were found in EST, proteomic, or the current guinea pig genome data.

NEIBank: genomics and bioinformatics resources for vision research.

NEIBank is an integrated resource for genomics and bioinformatics in vision research. It includes expressed sequence tag (EST) data and sequence-verified cDNA clones for multiple eye tissues of several species, web-based access to human eye-specific SAGE data through EyeSAGE, and comprehensive, annotated databases of known human eye disease genes and candidate disease gene loci. All expression- and disease-related data are integrated in EyeBrowse, an eye-centric genome browser. NEIBank provides a comprehensive overview of current knowledge of the transcriptional repertoires of eye tissues and their relation to pathology.

The lacrimal gland transcriptome is an unusually rich source of rare and poorly characterized gene transcripts.

PURPOSE: To sequence and comprehensively analyze human and mouse lacrimal gland transcriptomes as part of the NEIBank project. METHODS: cDNA libraries generated from normal human and mouse lacrimal glands were sequenced and analyzed by PHRED, RepeatMasker, BLAST, and GRIST. Human "lacrimal-preferred genes" and putative gene regulatory elements were respectively identified in UniGene and ConSite, and gene clustering was analyzed by chromosomal mapping. "Hypothetical proteins," identified by keyword search, were verified by genomic alignment and queried in the Conserved Domain database and GEO Profiles. RESULTS: The top six transcripts in human and mouse differed, revealing a previously unappreciated molecular divergence. The human transcriptome is enriched with transcripts from 29 lacrimal-preferred genes and a content of poorly characterized hypothetical proteins, proportionally greater than in all other tissues. Only 45% of lacrimal preferred, but 71% of hypotheticals, have mouse orthologs. Many of the latter display apparently altered cancer expression in the CGAP SAGE library collection-often in keeping with predicted WD40, protein kinase, Src homology 2 and 3, RhoGEF, and pleckstrin homology domains involved in cell signaling. At the genomic level, lacrimal-expressed genes show some evidence of clustering, particularly on human chromosomes 9 and 12. Binding sites for TFAP2A, FOXC1, and other transcription factors are predicted. CONCLUSIONS: Interspecies divergence cautions against use of mouse models of human dry eye syndromes. Lacrimal preferred and hypothetical proteins, gene clustering, and putative gene regulatory elements together provide new clues for a molecular understanding of lacrimal gland function and mechanisms of coordinated tissue-specific transcriptional regulation.

Expressed sequence tag analysis of zebrafish eye tissues for NEIBank.

PURPOSE: To characterize gene expression patterns in various tissues of the zebrafish (Danio rerio) eye and identify zebrafish orthologs of human genes by expressed sequence tag (EST) analysis for NEIBank. METHODS: mRNA was extracted from adult zebrafish eye tissues, including lenses, anterior segments (minus lens), retinas, posterior segments lacking retinas, and whole eyes. Five different cDNA libraries were constructed in the pCMVSport6 vector. Approximately 4,000 clones from each library were sequenced and analyzed using various bioinformatics programs. RESULTS: The analysis yielded approximately 2,500 different gene clusters for each library. Combining data from the five libraries produced 10,392 unique gene clusters. GenBank accession numbers were identified for 37.6% (3,906) of the total gene clusters in the combined libraries and approximately 50% were linked to Unigene clusters in the current database. Several new crystallin genes, including two gammaN-crystallins, and a second major intrinsic protein (MIP) were identified in the lens library. In addition, a zebrafish homolog of cochlin (COCH), a gene that may play a role in the pathogenesis of human glaucoma, was identified in the anterior segment library. Surprisingly, no clear ortholog of the major retinal transcription factor Nrl was identified. CONCLUSIONS: The zebrafish eye tissue cDNA libraries are a useful resource for comparative gene expression analysis. These libraries will complement the cDNA libraries made for the Zebrafish Gene Collection (ZGC) and provide an additional source for gene identification and characterization in the vertebrate eye.

Pycortex: an interactive surface visualizer for fMRI.

Surface visualizations of fMRI provide a comprehensive view of cortical activity. However, surface visualizations are difficult to generate and most common visualization techniques rely on unnecessary interpolation which limits the fidelity of the resulting maps. Furthermore, it is difficult to understand the relationship between flattened cortical surfaces and the underlying 3D anatomy using tools available currently. To address these problems we have developed pycortex, a Python toolbox for interactive surface mapping and visualization. Pycortex exploits the power of modern graphics cards to sample volumetric data on a per-pixel basis, allowing dense and accurate mapping of the voxel grid across the surface. Anatomical and functional information can be projected onto the cortical surface. The surface can be inflated and flattened interactively, aiding interpretation of the correspondence between the anatomical surface and the flattened cortical sheet. The output of pycortex can be viewed using WebGL, a technology compatible with modern web browsers. This allows complex fMRI surface maps to be distributed broadly online without requiring installation of complex software.

EST analysis of mouse retina and RPE/choroid cDNA libraries.

PURPOSE: cDNA libraries from the mouse retina have recently been reported, but no well characterized library from the retinal pigment epithelium (RPE) or choroid of the mouse has yet appeared in the literature. To complement these libraries and to provide the first mouse RPE/choroid library, we used freshly dissected tissue from adult C57BL/6J mice to construct new retina and RPE/choroid libraries. METHODS: Eyes from 100 six to eight week old C57BL/6J mice were dissected in groups of 10. The whole retina and RPE/choroid were isolated individually and then homogenized before RNA isolation. Over 5000 clones each were sequenced from the unamplified and un-normalized retina and RPE/choroid libraries. All sequences were analyzed using GRIST (GRouping and Identification of Sequence Tags), a bioinformatics program for gene identification and clustering. RESULTS: The RPE/choroid library contained 3145 clusters with 76% of the clusters representing single clones. Nearly 87% of the clusters corresponded to named genes in GenBank, and 8% of the RPE clusters remain unidentified. The retina library contained 3190 clusters of which 78% represented only one clone. Approximately 85% of the clusters matched sequences in GenBank, and 9% of the clusters remain unidentified. The clones most abundant in each library were all well-known sequences and both libraries contained a number of tissue specific or tissue-enhanced genes. CONCLUSIONS: These new libraries should provide a valuable resource for gene discovery and cDNAs for expression analysis and functional studies.

Grouping and identification of sequence tags (GRIST): bioinformatics tools for the NEIBank database.

NEIBank is a project to develop and organize genomics and bioinformatics resources for the eye. As part of this effort, tools have been developed for bioinformatics analysis and web based display of data from expressed sequence tag (EST) analyses. EST sequences are identified and formed into groups or clusters representing related transcripts from the same gene. This is carried out by a rules-based procedure called GRIST (GRouping and Identification of Sequence Tags) that uses sequence match parameters derived from BLAST programs. Linked procedures are used to eliminate non-mRNA contaminants. All data are assembled in a relational database and assembled for display as web pages with annotations and links to other informatics resources. Genome projects generate huge amounts of data that need to be classified and organized to become easily accessible to the research community. GRIST provides a useful tool for assembling and displaying the results of EST analyses. The NEIBank web site contains a growing set of pages cataloging the known transcriptional repertoire of eye tissues, derived from new NEIBank cDNA libraries and from eye-related data deposited in the dbEST section of GenBank.

A specific missense mutation in GTF2I occurs at high frequency in thymic epithelial tumors.

We analyzed 28 thymic epithelial tumors (TETs) using next-generation sequencing and identified a missense mutation (chromosome 7 c.74146970T>A) in GTF2I at high frequency in type A thymomas, a relatively indolent subtype. In a series of 274 TETs, we detected the GTF2I mutation in 82% of type A and 74% of type AB thymomas but rarely in the aggressive subtypes, where recurrent mutations of known cancer genes have been identified. Therefore, GTF2I mutation correlated with better survival. GTF2I ß and δ isoforms were expressed in TETs, and both mutant isoforms were able to stimulate cell proliferation in vitro. Thymic carcinomas carried a higher number of mutations than thymomas (average of 43.5 and 18.4, respectively). Notably, we identified recurrent mutations of known cancer genes, including TP53, CYLD, CDKN2A, BAP1 and PBRM1, in thymic carcinomas. These findings will complement the diagnostic assessment of these tumors and also facilitate development of a molecular classification and assessment of prognosis and treatment strategies.

Whole genome and transcriptome sequencing of a B3 thymoma.

Molecular pathology of thymomas is poorly understood. Genomic aberrations are frequently identified in tumors but no extensive sequencing has been reported in thymomas. Here we present the first comprehensive view of a B3 thymoma at whole genome and transcriptome levels. A 55-year-old Caucasian female underwent complete resection of a stage IVA B3 thymoma. RNA and DNA were extracted from a snap frozen tumor sample with a fraction of cancer cells over 80%. We performed array comparative genomic hybridization using Agilent platform, transcriptome sequencing using HiSeq 2000 (Illumina) and whole genome sequencing using Complete Genomics Inc platform. Whole genome sequencing determined, in tumor and normal, the sequence of both alleles in more than 95% of the reference genome (NCBI Build 37). Copy number (CN) aberrations were comparable with those previously described for B3 thymomas, with CN gain of chromosome 1q, 5, 7 and X and CN loss of 3p, 6, 11q42.2-qter and q13. One translocation t(11;X) was identified by whole genome sequencing and confirmed by PCR and Sanger sequencing. Ten single nucleotide variations (SNVs) and 2 insertion/deletions (INDELs) were identified; these mutations resulted in non-synonymous amino acid changes or affected splicing sites. The lack of common cancer-associated mutations in this patient suggests that thymomas may evolve through mechanisms distinctive from other tumor types, and supports the rationale for additional high-throughput sequencing screens to better understand the somatic genetic architecture of thymoma.

Inverse folding of RNA pseudoknot structures.

BACKGROUND: RNA exhibits a variety of structural configurations. Here we consider a structure to be tantamount to the noncrossing Watson-Crick and G-U-base pairings (secondary structure) and additional cross-serial base pairs. These interactions are called pseudoknots and are observed across the whole spectrum of RNA functionalities. In the context of studying natural RNA structures, searching for new ribozymes and designing artificial RNA, it is of interest to find RNA sequences folding into a specific structure and to analyze their induced neutral networks. Since the established inverse folding algorithms, RNAinverse, RNA-SSD as well as INFO-RNA are limited to RNA secondary structures, we present in this paper the inverse folding algorithm Inv which can deal with 3-noncrossing, canonical pseudoknot structures. RESULTS: In this paper we present the inverse folding algorithm Inv. We give a detailed analysis of Inv, including pseudocodes. We show that Inv allows to design in particular 3-noncrossing nonplanar RNA pseudoknot 3-noncrossing RNA structures-a class which is difficult to construct via dynamic programming routines. Inv is freely available at http://www.combinatorics.cn/cbpc/inv.html. CONCLUSIONS: The algorithm Inv extends inverse folding capabilities to RNA pseudoknot structures. In comparison with RNAinverse it uses new ideas, for instance by considering sets of competing structures. As a result, Inv is not only able to find novel sequences even for RNA secondary structures, it does so in the context of competing structures that potentially exhibit cross-serial interactions.

Gene expression profile of the mouse organ of Corti at the onset of hearing.

We describe the generation of an expressed sequence tag (EST) database of the mouse organ of Corti (OC). Over 20,000 independent clones were isolated, analyzed, and grouped into 8690 unique gene clusters. A large pool of novel genes unique to the OC was identified. Sequence alignments frequently revealed alternatively spliced forms of known genes potentially relevant in the OC function. We have also electronically mapped a subset of OC mouse ESTs to several syntenic regions associated with human autosomal and recessive deafness, which may prove useful for the identification of new positional candidates for these human diseases. The EST dataset is available as an interactive Web-based public database at. This resource provides both a view of the profile of gene expression in the OC at the onset of hearing and a tool to identify novel genes of importance in hearing.