Deep Learning Assisted Imaging Methods to Facilitate Access to Ophthalmic Telepathology.

Purpose: To investigate the use of super-resolution imaging techniques to enable telepathology using low-cost commercial cameras. Design: Experimental study. Participants: A total of 139 ophthalmic pathology slides obtained from the Ophthalmic Pathology service at the University of California, Irvine. Methods: Denoising Diffusion Probabilistic Model (DDPM) was developed to predict super-resolution pathology slide images from low-resolution inputs. The model was pretrained using 150 000 images randomly sampled from the ImageNet dataset. Patch aggregation was used to generate large images with DDPM. The performance of DDPM was evaluated against that of generative adversarial networks (GANs) and Robust UNet, which were also trained on the same dataset. Main Outcome Measures: The performance of models trained to generate super-resolution output images from low-resolution input images can be evaluated by using the mean squared error (MSE) and Structural Similarity Index Measure (SSIM), as well as subjective grades provided by expert pathologist graders. Results: In total, our study included 110 training images, 9 validation images, and 20 testing images. The objective performance scores were averaged over patches generated from 20 test images. The DDPM-based approach with pretraining produced the best results, with an MSE score of 1.35e-5 and an SSIM score of 0.8987. A qualitative analysis of super-resolution images was conducted by expert 3 pathologists and 1 expert ophthalmic microscopist, and the average accuracy of identifying the correct ground truth images ranged from 25% to 70% (with an average accuracy of 46.5%) for widefield images and 25% to 60% (with an average accuracy of 38.25%) for individual patches. Conclusions: The DDPM-based approach with pretraining is assessed to be effective at super-resolution prediction for ophthalmic pathology slides both in terms of objective and subjective measures. The proposed methodology is expected to decrease the reliance on costly slide scanners for acquiring high-quality pathology slide images, while also streamlining clinical workflow and expanding the scope of ophthalmic telepathology. Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Vitreoretinal Surgical Instrument Tracking in Three Dimensions Using Deep Learning.

Purpose: To evaluate the potential for artificial intelligence-based video analysis to determine surgical instrument characteristics when moving in the three-dimensional vitreous space. Methods: We designed and manufactured a model eye in which we recorded choreographed videos of many surgical instruments moving throughout the eye. We labeled each frame of the videos to describe the surgical tool characteristics: tool type, location, depth, and insertional laterality. We trained two different deep learning models to predict each of the tool characteristics and evaluated model performances on a subset of images. Results: The accuracy of the classification model on the training set is 84% for the x-y region, 97% for depth, 100% for instrument type, and 100% for laterality of insertion. The accuracy of the classification model on the validation dataset is 83% for the x-y region, 96% for depth, 100% for instrument type, and 100% for laterality of insertion. The close-up detection model performs at 67 frames per second, with precision for most instruments higher than 75%, achieving a mean average precision of 79.3%. Conclusions: We demonstrated that trained models can track surgical instrument movement in three-dimensional space and determine instrument depth, tip location, instrument insertional laterality, and instrument type. Model performance is nearly instantaneous and justifies further investigation into application to real-world surgical videos. Translational Relevance: Deep learning offers the potential for software-based safety feedback mechanisms during surgery or the ability to extract metrics of surgical technique that can direct research to optimize surgical outcomes.

Deep learning to enable color vision in the dark.

Humans perceive light in the visible spectrum (400-700 nm). Some night vision systems use infrared light that is not perceptible to humans and the images rendered are transposed to a digital display presenting a monochromatic image in the visible spectrum. We sought to develop an imaging algorithm powered by optimized deep learning architectures whereby infrared spectral illumination of a scene could be used to predict a visible spectrum rendering of the scene as if it were perceived by a human with visible spectrum light. This would make it possible to digitally render a visible spectrum scene to humans when they are otherwise in complete "darkness" and only illuminated with infrared light. To achieve this goal, we used a monochromatic camera sensitive to visible and near infrared light to acquire an image dataset of printed images of faces under multispectral illumination spanning standard visible red (604 nm), green (529 nm) and blue (447 nm) as well as infrared wavelengths (718, 777, and 807 nm). We then optimized a convolutional neural network with a U-Net-like architecture to predict visible spectrum images from only near-infrared images. This study serves as a first step towards predicting human visible spectrum scenes from imperceptible near-infrared illumination. Further work can profoundly contribute to a variety of applications including night vision and studies of biological samples sensitive to visible light.

The DNA modification N6-methyl-2'-deoxyadenosine (m6dA) drives activity-induced gene expression and is required for fear extinction.

DNA modification is known to regulate experience-dependent gene expression. However, beyond cytosine methylation and its oxidated derivatives, very little is known about the functional importance of chemical modifications on other nucleobases in the brain. Here we report that in adult mice trained in fear extinction, the DNA modification N6-methyl-2'-deoxyadenosine (m6dA) accumulates along promoters and coding sequences in activated prefrontal cortical neurons. The deposition of m6dA is associated with increased genome-wide occupancy of the mammalian m6dA methyltransferase, N6amt1, and this correlates with extinction-induced gene expression. The accumulation of m6dA is associated with transcriptional activation at the brain-derived neurotrophic factor (Bdnf) P4 promoter, which is required for Bdnf exon IV messenger RNA expression and for the extinction of conditioned fear. These results expand the scope of DNA modifications in the adult brain and highlight changes in m6dA as an epigenetic mechanism associated with activity-induced gene expression and the formation of fear extinction memory.

Atmospheric Oxidation of Squalene: Molecular Study Using COBRA Modeling and High-Resolution Mass Spectrometry.

Squalene is a major component of skin and plant surface lipids and is known to be present at high concentrations in indoor dust. Its high reactivity toward ozone makes it an important ozone sink and a natural protectant against atmospheric oxidizing agents. While the volatile products of squalene ozonolysis are known, the condensed-phase products have not been characterized. We present an analysis of condensed-phase products resulting from an extensive oxidation of squalene by ozone probed by electrospray ionization (ESI) high-resolution mass spectrometry (HR-MS). A complex distribution of nearly 1300 peaks assignable to molecular formulas is observed in direct infusion positive ion mode ESI mass spectra. The distribution of peaks in the mass spectra suggests that there are extensive cross-coupling reactions between hydroxy-carbonyl products of squalene ozonolysis. To get additional insights into the mechanism, we apply a Computational Brewing Application (COBRA) to simulate the oxidation of squalene in the presence of ozone, and compare predicted results with those observed by the HR-MS experiments. The system predicts over one billion molecular structures between 0 and 1450 Da, which correspond to about 27 000 distinct elemental formulas. Over 83% of the squalene oxidation products inferred from the mass spectrometry data are matched by the simulation. The simulation indicates a prevalence of peroxy groups, with hydroxyl and ether groups being the second-most important O-containing functional groups formed during squalene oxidation. These highly oxidized products of squalene ozonolysis may accumulate on indoor dust and surfaces and contribute to their redox capacity.

A GRHL3-regulated repair pathway suppresses immune-mediated epidermal hyperplasia.

Dermal infiltration of T cells is an important step in the onset and progression of immune-mediated skin diseases such as psoriasis; however, it is not known whether epidermal factors play a primary role in the development of these diseases. Here, we determined that the prodifferentiation transcription factor grainyhead-like 3 (GRHL3), which is essential during epidermal development, is dispensable for adult skin homeostasis, but required for barrier repair after adult epidermal injury. Consistent with activation of a GRHL3-regulated repair pathway in psoriasis, we found that GRHL3 is upregulated in lesional skin and binds known epidermal differentiation gene targets. Using an imiquimod-induced model of immune-mediated epidermal hyperplasia, we found that mice lacking GRHL3 have an exacerbated epidermal damage response, greater sensitivity to disease induction, delayed resolution of epidermal lesions, and resistance to anti-IL-22 therapy compared with WT animals. ChIP-Seq and gene expression profiling of murine skin revealed that while GRHL3 regulates differentiation pathways both during development and during repair from immune-mediated damage, it targets distinct sets of genes in the 2 processes. In particular, GRHL3 suppressed a number of alarmin and other proinflammatory genes after immune injury. This study identifies a GRHL3-regulated epidermal barrier repair pathway that suppresses disease initiation and helps resolve existing lesions in immune-mediated epidermal hyperplasia.

Visual adaptation and novelty responses in the superior colliculus.

The brain's ability to ignore repeating, often redundant, information while enhancing novel information processing is paramount to survival. When stimuli are repeatedly presented, the response of visually sensitive neurons decreases in magnitude, that is, neurons adapt or habituate, although the mechanism is not yet known. We monitored the activity of visual neurons in the superior colliculus (SC) of rhesus monkeys who actively fixated while repeated visual events were presented. We dissociated adaptation from habituation as mechanisms of the response decrement by using a Bayesian model of adaptation, and by employing a paradigm including rare trials that included an oddball stimulus that was either brighter or dimmer. If the mechanism is adaptation, response recovery should be seen only for the brighter stimulus; if the mechanism is habituation, response recovery ('dishabituation') should be seen for both the brighter and dimmer stimuli. We observed a reduction in the magnitude of the initial transient response and an increase in response onset latency with stimulus repetition for all visually responsive neurons in the SC. Response decrement was successfully captured by the adaptation model, which also predicted the effects of presentation rate and rare luminance changes. However, in a subset of neurons with sustained activity in response to visual stimuli, a novelty signal akin to dishabituation was observed late in the visual response profile for both brighter and dimmer stimuli, and was not captured by the model. This suggests that SC neurons integrate both rapidly discounted information about repeating stimuli and novelty information about oddball events, to support efficient selection in a cluttered dynamic world.

Large scale study of multiple-molecule queries.

BACKGROUND: In ligand-based screening, as well as in other chemoinformatics applications, one seeks to effectively search large repositories of molecules in order to retrieve molecules that are similar typically to a single molecule lead. However, in some case, multiple molecules from the same family are available to seed the query and search for other members of the same family.Multiple-molecule query methods have been less studied than single-molecule query methods. Furthermore, the previous studies have relied on proprietary data and sometimes have not used proper cross-validation methods to assess the results. In contrast, here we develop and compare multiple-molecule query methods using several large publicly available data sets and background. We also create a framework based on a strict cross-validation protocol to allow unbiased benchmarking for direct comparison in future studies across several performance metrics. RESULTS: Fourteen different multiple-molecule query methods were defined and benchmarked using: (1) 41 publicly available data sets of related molecules with similar biological activity; and (2) publicly available background data sets consisting of up to 175,000 molecules randomly extracted from the ChemDB database and other sources. Eight of the fourteen methods were parameter free, and six of them fit one or two free parameters to the data using a careful cross-validation protocol. All the methods were assessed and compared for their ability to retrieve members of the same family against the background data set by using several performance metrics including the Area Under the Accumulation Curve (AUAC), Area Under the Curve (AUC), F1-measure, and BEDROC metrics.Consistent with the previous literature, the best parameter-free methods are the MAX-SIM and MIN-RANK methods, which score a molecule to a family by the maximum similarity, or minimum ranking, obtained across the family. One new parameterized method introduced in this study and two previously defined methods, the Exponential Tanimoto Discriminant (ETD), the Tanimoto Power Discriminant (TPD), and the Binary Kernel Discriminant (BKD), outperform most other methods but are more complex, requiring one or two parameters to be fit to the data. CONCLUSION: Fourteen methods for multiple-molecule querying of chemical databases, including novel methods, (ETD) and (TPD), are validated using publicly available data sets, standard cross-validation protocols, and established metrics. The best results are obtained with ETD, TPD, BKD, MAX-SIM, and MIN-RANK. These results can be replicated and compared with the results of future studies using data freely downloadable from http://cdb.ics.uci.edu/.

Phylogenetic analysis, genome evolution and the rate of gene gain in the Herpesviridae.

We used complete sequence data from 30 complete Herpesviridae genomes to investigate phylogenetic relationships and patterns of genome evolution. The approach was to identify orthologous gene clusters among taxa and to generate a genomic matrix of gene content. We identified 17 genes with homologs in all 30 taxa and concatenated a subset of 10 of these genes for phylogenetic inference. We also constructed phylogenetic trees on the basis of gene content data. The amino acid and gene content phylogenies were largely concordant, but the amino acid data had much higher internal support. We mapped gene gain events onto the phylogenetic tree by assuming that genes were gained only once during the evolution of herpesviruses. Thirty genes were inferred to be present in the ancestor of all herpesvirus, a number smaller than previously hypothesized. Few genes of recent origin within herpesviruses could be identified as originating from transfer between virus and vertebrate hosts. Inferred rates of gene gain were heterogeneous, with both taxonomic and temporal biases. Nonetheless, the average rate of gene gain was approximately 3.5 x 10(-7) genes gained per year, which is an order of magnitude higher than the nucleotide mutation rate for these large DNA viruses.

Genomic splice-site analysis reveals frequent alternative splicing close to the dominant splice site.

Alternative pre-mRNA splicing may be the most efficient and widespread mechanism to generate multiple protein isoforms from single genes. Here, we describe the genomic analysis of one of the most frequent types of alternative pre-mRNA splicing, alternative 5'- and 3'-splice-site selection. Using an EST-based alternative splicing database recording >47,000 alternative splicing events, we determined the frequency and location of alternative 5'- and 3'-splice sites within the human genome. The most common alternative splice sites used in the human genome are located within 6 nucleotides (nt) of the dominant splice site. We show that the EST database overrepresents alternative splicing events that maintain the reading frame, thus supporting the concept that RNA quality-control steps ensure that mRNAs that encode for potentially harmful protein products are destroyed and do not serve as templates for translation. The most frequent location for alternative 5'-splice sites is 4 nt upstream or downstream from the dominant splice site. Sequence analysis suggests that this preference is a consequence of the U1 snRNP binding sequence at the 5'-splice site, which frequently contains a GU dinucleotide 4 nt downstream from the dominant splice site. Surprisingly, approximately 50% of duplicated 3'-YAG splice junctions are subject to alternative splicing. This high probability of alternative 3'-splice-site activation in close proximity of the dominant 3'-splice site suggests that the second step of the splicing may be prone to violate splicing fidelity.

The architecture of pre-mRNAs affects mechanisms of splice-site pairing.

The exon/intron architecture of genes determines whether components of the spliceosome recognize splice sites across the intron or across the exon. Using in vitro splicing assays, we demonstrate that splice-site recognition across introns ceases when intron size is between 200 and 250 nucleotides. Beyond this threshold, splice sites are recognized across the exon. Splice-site recognition across the intron is significantly more efficient than splice-site recognition across the exon, resulting in enhanced inclusion of exons with weak splice sites. Thus, intron size can profoundly influence the likelihood that an exon is constitutively or alternatively spliced. An EST-based alternative-splicing database was used to determine whether the exon/intron architecture influences the probability of alternative splicing in the Drosophila and human genomes. Drosophila exons flanked by long introns display an up to 90-fold-higher probability of being alternatively spliced compared with exons flanked by two short introns, demonstrating that the exon/intron architecture in Drosophila is a major determinant in governing the frequency of alternative splicing. Exon skipping is also more likely to occur when exons are flanked by long introns in the human genome. Interestingly, experimental and computational analyses show that the length of the upstream intron is more influential in inducing alternative splicing than is the length of the downstream intron. We conclude that the size and location of the flanking introns control the mechanism of splice-site recognition and influence the frequency and the type of alternative splicing that a pre-mRNA transcript undergoes.

Polymodal sensory function of the Caenorhabditis elegans OCR-2 channel arises from distinct intrinsic determinants within the protein and is selectively conserved in mammalian TRPV proteins.

Caenorhabditis elegans OCR-2 (OSM-9 and capsaicin receptor-related) is a TRPV (vanilloid subfamily of transient receptor potential channel) protein that regulates serotonin (5-HT) biosynthesis in chemosensory neurons and also mediates olfactory and osmotic sensation. Here, we identify the molecular basis for the polymodal function of OCR-2 in its native cellular environment. We show that OCR-2 function in 5-HT production and osmotic sensing is governed by its N-terminal region upstream of the ankyrin repeats domain, but the diacetyl sensitivity is mediated by independent mechanisms. The ocr-2(yz5) mutation results in a glycine-to-glutamate substitution (G36E) within the N-terminal region. The G36E substitution causes dramatic downregulation of 5-HT synthesis in the ADF neurons, eliminates osmosensation mediated by the ASH neurons, but does not affect the response to the odorant diacetyl mediated by the AWA neurons. Conversely, wild-type sequence of the N-terminal segment confers osmotic sensitivity and upregulation of 5-HT production to a normally insensitive C. elegans homolog, OCR-4, but this chimeric channel does not respond to diacetyl stimuli. Furthermore, expression of either the mouse or human TRPV2 gene under the ocr-2 promoter can substantially restore 5-HT biosynthesis in ocr-2-null mutants but cannot improve the deficits in osmotic or olfactory sensation, suggesting that TRPV2 can substitute for the role of OCR-2 only in serotonergic neurons. Thus, different sensory functions of OCR-2 arise from separable intrinsic determinants, and specific functional properties of TRPV channel proteins may be selectively conserved across phyla.

Extensive gene gain associated with adaptive evolution of poxviruses.

Previous studies of genome evolution usually have involved one or two genomes and have thus been limited in their ability to detect the direction and rate of evolutionary change. Here, we use complete genome data from 20 poxvirus genomes to build a robust phylogeny of the Poxviridae and to study patterns of genome evolution. We show that, although there has been little gene order evolution, there are substantial differences between poxviruses in terms of genome content. Furthermore, we show that the rate of gene acquisition is not constant over time and that it has increased in the orthopox lineage (which includes smallpox and vaccinia). We also tested for positive selection on 204 groups of genes and show that a disproportionately high proportion of genes in the orthopox clade are under positive selection. The association of an increased rate of gene gain and positive selection is indicative of adaptive genome evolution. Many of the genes involved in these processes are likely to be associated with host-parasite coevolution.

Comprehensive mutagenesis of the C-terminal domain of the M13 gene-3 minor coat protein: the requirements for assembly into the bacteriophage particle.

Filamentous bacteriophage assemble at the host membrane in a non-lytic process; the gene-3 minor coat protein (P3) is required for release from the membrane and subsequently, for recognition and infection of a new host. P3 contains at least three distinct domains: two N-terminal domains that mediate host recognition and infection, and a C-terminal domain (P3-C) that is required for release from the host cell following phage assembly and contributes to the structural stability of the phage particle. A comprehensive mutational analysis of the 150 residue P3-C revealed that only 24 side-chains, located within the last 70 residues of sequence, were necessary for efficient incorporation into a wild-type coat. The results reveal that the requirements for the assembly of P3 into the phage particle are quite lax and involve only a few key side-chains. These findings shed light on the functional and structural requirements for filamentous phage assembly, and they may provide guidelines for the engineering of improved coat proteins as scaffolds for phage display technology.