Performance of an artificial intelligence algorithm for reporting urine cytopathology.

BACKGROUND: Unlike Papanicolaou tests, there are no commercially available computer-assisted automated screening systems for urine specimens. Despite The Paris System for Reporting Urinary Cytology, there still is poor interobserver agreement with urine cytology and many cases in which a definitive diagnosis cannot be made. In the current study, the authors have reported on the development of an image algorithm that applies computational methods to digitized liquid-based urine cytology slides. METHODS: A total of 2405 archival ThinPrep glass slides, including voided and instrumented urine cytology cases, were digitized. A deep learning computational pipeline with multiple tiers of convolutional neural network models was developed for processing whole slide images (WSIs) and predicting diagnoses. The algorithm was validated using a separate test data set comprised of consecutive cases encountered in routine clinical practice. RESULTS: There were 1.9 million urothelial cells analyzed. An average of 5400 urothelial cells were identified in each WSI. The algorithm achieved an area under the curve of 0.88 (95% CI, 0.83-0.93). Using the optimal operating point, the algorithm's sensitivity was 79.5% (95% CI, 64.7%-90.2%) and the specificity was 84.5% (95% CI, 81.6%-87.1%) for high-grade urothelial carcinoma. CONCLUSIONS: The authors successfully developed a computational algorithm capable of accurately analyzing WSIs of urine cytology cases. Compared with prior studies, this effort used a much larger data set, exploited whole slide-level and not just cell-level features, and used a cell gallery to display the algorithm's output for easy end-user review. This algorithm provides computer-assisted interpretation of urine cytology cases, akin to the machine learning technology currently used for automated Papanicolaou test screening.

hgvs: A Python package for manipulating sequence variants using HGVS nomenclature: 2018 Update.

The Human Genome Variation Society (HGVS) nomenclature guidelines encourage the accurate and standard description of DNA, RNA, and protein sequence variants in public variant databases and the scientific literature. Inconsistent application of the HGVS guidelines can lead to misinterpretation of variants in clinical settings. Reliable software tools are essential to ensure consistent application of the HGVS guidelines when reporting and interpreting variants. We present the hgvs Python package, a comprehensive tool for manipulating sequence variants according to the HGVS nomenclature guidelines. Distinguishing features of the hgvs package include: (1) parsing, formatting, validating, and normalizing variants on genome, transcript, and protein sequences; (2) projecting variants between aligned sequences, including those with gapped alignments; (3) flexible installation using remote or local data (fully local installations eliminate network dependencies); (4) extensive automated tests; and (5) open source development by a community from eight organizations worldwide. This report summarizes recent and significant updates to the hgvs package since its original release in 2014, and presents results of extensive validation using clinical relevant variants from ClinVar and HGMD.

Clinical Implementation and Validation of Automated Human Genome Variation Society (HGVS) Nomenclature System for Next-Generation Sequencing-Based Assays for Cancer.

Human Genome Variation Society (HGVS) nomenclature is a de facto clinical standard for reporting DNA sequence variants. With increasing use of high-throughput sequencing, manual generation of HGVS nomenclatures for all variants is impractical and error-prone. It is therefore beneficial to include one or more HGVS generator tools in next-generation sequencing (NGS) bioinformatics pipelines to enable automated, consistent, and accurate generation of HGVS nomenclature after appropriate validation. The authors implemented an HGVS nomenclature tool, the hgvs package, by integrating it into their custom-developed NGS variant management and reporting software. Use of Docker containers provided a strategic advantage to the integration process. Clinical implementation of the hgvs package was validated using a cohort of 330 variants that appropriately represented cancer-related genes and clinically important variant types. The hgvs package was able to generate HGVS-compliant variant nomenclature (both c. and p.) for 308 of the 330 (93.3%) variants, including all those in the coding and untranslated regions, and 32 of 35 (91.4%) in the consensus splice site region. Discrepant HGVS nomenclature involved variants in the intronic (16 of 40) and consensus splice site (3 of 35) regions with repeat sequences. Overall, implementation of the hgvs package in the clinical NGS workflow improved consistency and accuracy of reporting HGVS nomenclature.

Transmembrane helix hydrophobicity is an energetic barrier during the retrotranslocation of integral membrane ERAD substrates.

Integral membrane proteins fold inefficiently and are susceptible to turnover via the endoplasmic reticulum-associated degradation (ERAD) pathway. During ERAD, misfolded proteins are recognized by molecular chaperones, polyubiquitinated, and retrotranslocated to the cytoplasm for proteasomal degradation. Although many aspects of this pathway are defined, how transmembrane helices (TMHs) are removed from the membrane and into the cytoplasm before degradation is poorly understood. In this study, we asked whether the hydrophobic character of a TMH acts as an energetic barrier to retrotranslocation. To this end, we designed a dual-pass model ERAD substrate, Chimera A*, which contains the cytoplasmic misfolded domain from a characterized ERAD substrate, Sterile 6* (Ste6p*). We found that the degradation requirements for Chimera A* and Ste6p* are similar, but Chimera A* was retrotranslocated more efficiently than Ste6p* in an in vitro assay in which retrotranslocation can be quantified. We then constructed a series of Chimera A* variants containing synthetic TMHs with a range of ΔG values for membrane insertion. TMH hydrophobicity correlated inversely with retrotranslocation efficiency, and in all cases, retrotranslocation remained Cdc48p dependent. These findings provide insight into the energetic restrictions on the retrotranslocation reaction, as well as a new computational approach to predict retrotranslocation efficiency.

Endoscopic ablation is a cost-effective cancer preventative therapy in patients with Barrett's esophagus who have elevated genomic instability.

BACKGROUND: The surveillance of patients with nondysplastic Barrett's esophagus (NDBE) has a high cost and is of limited effectiveness in preventing esophageal adenocarcinoma (EAC). Ablation for NDBE remains expensive and controversial. Biomarkers of genomic instability have shown promise in identifying patients with NDBE at high risk for progression to EAC. Here, we evaluate the cost-effectiveness of using such biomarkers to stratify patients with NDBE by risk for EAC and, subsequently, the cost-effectiveness of ablative therapy. METHODS: A Markov decision tree was used to evaluate four strategies in a hypothetical cohort of 50-year old patients with NDBE over their lifetime: strategy I, natural history without surveillance; strategy II, surveillance per current guidelines; strategy III, ablation for all patients; strategy IV, risk stratification with use of a biomarker panel to assess genomic instability (i. e., mutational load [ML]). Patients with no ML underwent minimal surveillance, patients with low ML underwent standard surveillance, and patients with high ML underwent ablation. The incremental cost-effectiveness ratio (ICER) and incremental net health benefit (INHB) were assessed. RESULTS: Strategy IV provided the best values for quality-adjusted life years (QALYs), ICER, and INHB in comparison with strategies II and III. RESULTS were robust in sensitivity analysis. In a Monte Carlo analysis, the relative risk for the development of cancer in the patients managed with strategy IV was decreased. Critical determinants of strategy IV cost-effectiveness were the complete response rate, cost of ablation, and surveillance interval in patients with no ML. CONCLUSION: The use of ML to stratify patients with NDBE by risk was the most cost-effective strategy for preventive EAC treatment. Targeting ablation toward patients with high ML presents an opportunity for a paradigm shift in the management of NDBE.

The Presence of Genetic Mutations at Key Loci Predicts Progression to Esophageal Adenocarcinoma in Barrett's Esophagus.

OBJECTIVES: Risk stratification in Barrett's esophagus (BE) is challenging. We evaluated the ability of a panel of genetic markers to predict progression to high-grade dysplasia (HGD) or esophageal adenocarcinoma (EAC). METHODS: In this case-control study, we assessed a measure of genetic instability, the mutational load (ML), in predicting progression to HGD or EAC. Cases had nondysplastic BE or low-grade dysplasia (LGD) at baseline and developed HGD/EAC ≥1 year later. Controls were matched 2:1, had nondysplastic BE or LGD, and no progression at follow-up. Formalin-fixed, paraffin-embedded tissue was microdissected for the epithelium. Loss of heterozygosity (LOH) and microsatellite instability (MSI) were assessed. ML was calculated from derangements in 10 genomic loci. High-clonality LOH mutations were assigned a value of 1, low-clonality mutations were assigned a value of 0.5, and MSI 0.75 at the first loci, and 0.5 for additional loci. These values were summed to the ML. Receiver operator characteristic (ROC) curves were created. RESULTS: There were 69 patients (46 controls and 23 cases). Groups were similar in age, follow-up time, baseline histology, and the number of microdissected targets. Mean ML in pre-progression biopsies was higher in cases (2.21) than in controls (0.42; P<0.0001). Sensitivity was 100% at ML ≥0.5 and specificity was 96% at ML ≥1.5. Accuracy was highest at 89.9% for ML ≥1. ROC curves for ML ≥1 demonstrated an area under the curve (AUC) of 0.95. CONCLUSIONS: ML in pre-progression BE tissue predicts progression to HGD or EAC. Although further validation is necessary, ML may have utility as a biomarker in endoscopic surveillance of BE.

Continuum approaches to understanding ion and peptide interactions with the membrane.

Experimental and computational studies have shown that cellular membranes deform to stabilize the inclusion of transmembrane (TM) proteins harboring charge. Recent analysis suggests that membrane bending helps to expose charged and polar residues to the aqueous environment and polar head groups. We previously used elasticity theory to identify membrane distortions that minimize the insertion of charged TM peptides into the membrane. Here, we extend our work by showing that it also provides a novel, computationally efficient method for exploring the energetics of ion and small peptide penetration into membranes. First, we show that the continuum method accurately reproduces energy profiles and membrane shapes generated from molecular simulations of bare ion permeation at a fraction of the computational cost. Next, we demonstrate that the dependence of the ion insertion energy on the membrane thickness arises primarily from the elastic properties of the membrane. Moreover, the continuum model readily provides a free energy decomposition into components not easily determined from molecular dynamics. Finally, we show that the energetics of membrane deformation strongly depend on membrane patch size both for ions and peptides. This dependence is particularly strong for peptides based on simulations of a known amphipathic, membrane binding peptide from the human pathogen Toxoplasma gondii. In total, we address shortcomings and advantages that arise from using a variety of computational methods in distinct biological contexts.

Antibacterial membrane attack by a pore-forming intestinal C-type lectin.

Human body-surface epithelia coexist in close association with complex bacterial communities and are protected by a variety of antibacterial proteins. C-type lectins of the RegIII family are bactericidal proteins that limit direct contact between bacteria and the intestinal epithelium and thus promote tolerance to the intestinal microbiota. RegIII lectins recognize their bacterial targets by binding peptidoglycan carbohydrate, but the mechanism by which they kill bacteria is unknown. Here we elucidate the mechanistic basis for RegIII bactericidal activity. We show that human RegIIIα (also known as HIP/PAP) binds membrane phospholipids and kills bacteria by forming a hexameric membrane-permeabilizing oligomeric pore. We derive a three-dimensional model of the RegIIIα pore by docking the RegIIIα crystal structure into a cryo-electron microscopic map of the pore complex, and show that the model accords with experimentally determined properties of the pore. Lipopolysaccharide inhibits RegIIIα pore-forming activity, explaining why RegIIIα is bactericidal for Gram-positive but not Gram-negative bacteria. Our findings identify C-type lectins as mediators of membrane attack in the mucosal immune system, and provide detailed insight into an antibacterial mechanism that promotes mutualism with the resident microbiota.

Membrane bending is critical for the stability of voltage sensor segments in the membrane.

The interaction between membrane proteins and the surrounding membrane is becoming increasingly appreciated for its role in regulating protein function, protein localization, and membrane morphology. In particular, recent studies have suggested that membrane deformation is needed to stably accommodate proteins harboring charged amino acids in their transmembrane (TM) region, as it is energetically prohibitive to bury charge in the hydrophobic core of the bilayer. Unfortunately, current computational methods are poorly equipped for describing such deformations, as atomistic simulations are often too short to observe large-scale membrane reorganization and most continuum approaches assume a flat membrane. Previously, we developed a method that overcomes these shortcomings by using elasticity theory to characterize equilibrium membrane distortions in the presence of a TM protein, while using traditional continuum electrostatic and nonpolar energy models to determine the energy of the protein in the membrane. Here, we linked the elastostatics, electrostatics, and nonpolar numeric solvers to permit the calculation of energies for nontrivial membrane deformations. We then coupled this procedure to a robust search algorithm that identifies optimal membrane shapes for a TM protein of arbitrary chemical composition. This advance now permits us to explore a host of biological phenomena that were beyond the scope of our original method. We show that the energy required to embed charged residues in the membrane can be highly nonadditive, and our model provides a simple mechanical explanation for this nonadditivity. Our results also predict that isolated voltage sensor segments do not insert into rigid membranes, but membrane bending dramatically stabilizes these proteins in the bilayer despite their high charge content. Additionally, we use the model to explore hydrophobic mismatch with regard to nonpolar peptides and mechanosensitive channels. Our method is in quantitative agreement with molecular dynamics simulations at a tiny fraction of the computational cost.

APBSmem: a graphical interface for electrostatic calculations at the membrane.

Electrostatic forces are one of the primary determinants of molecular interactions. They help guide the folding of proteins, increase the binding of one protein to another and facilitate protein-DNA and protein-ligand binding. A popular method for computing the electrostatic properties of biological systems is to numerically solve the Poisson-Boltzmann (PB) equation, and there are several easy-to-use software packages available that solve the PB equation for soluble proteins. Here we present a freely available program, called APBSmem, for carrying out these calculations in the presence of a membrane. The Adaptive Poisson-Boltzmann Solver (APBS) is used as a back-end for solving the PB equation, and a Java-based graphical user interface (GUI) coordinates a set of routines that introduce the influence of the membrane, determine its placement relative to the protein, and set the membrane potential. The software Jmol is embedded in the GUI to visualize the protein inserted in the membrane before the calculation and the electrostatic potential after completing the computation. We expect that the ease with which the GUI allows one to carry out these calculations will make this software a useful resource for experimenters and computational researchers alike. Three examples of membrane protein electrostatic calculations are carried out to illustrate how to use APBSmem and to highlight the different quantities of interest that can be calculated.