Deep phenotyping has been defined as the precise and comprehensive analysis of phenotypic abnormalities in which the individual components of the phenotype are observed and described. The three components of the Human Phenotype Ontology (HPO; www.human-phenotype-ontology.org) project are the phenotype vocabulary, disease-phenotype annotations and the algorithms that operate on these. These components are being used for computational deep phenotyping and precision medicine as well as integration of clinical data into translational research. The HPO is being increasingly adopted as a standard for phenotypic abnormalities by diverse groups such as international rare disease organizations, registries, clinical labs, biomedical resources, and clinical software tools and will thereby contribute toward nascent efforts at global data exchange for identifying disease etiologies. This update article reviews the progress of the HPO project since the debut Nucleic Acids Research database article in 2014, including specific areas of expansion such as common (complex) disease, new algorithms for phenotype driven genomic discovery and diagnostics, integration of cross-species mapping efforts with the Mammalian Phenotype Ontology, an improved quality control pipeline, and the addition of patient-friendly terminology.

Crohn’s disease (CD)–associated variants in the LRRK2 gene for risk (N2081D) and for protection (N551K) mediate shared effects in CD and Parkinson’s disease.

Standardization of immunophenotyping requires careful attention to reagents, sample handling, instrument setup, and data analysis, and is essential for successful cross-study and cross-center comparison of data. Experts developed five standardized, eight-color panels for identification of major immune cell subsets in peripheral blood. These were produced as pre-configured, lyophilized, reagents in 96-well plates. We present the results of a coordinated analysis of samples across nine laboratories using these panels with standardized operating procedures (SOPs). Manual gating was performed by each site and by a central site. Automated gating algorithms were developed and tested by the FlowCAP consortium. Centralized manual gating can reduce cross-center variability, and we sought to determine whether automated methods could streamline and standardize the analysis. Within-site variability was low in all experiments, but cross-site variability was lower when central analysis was performed in comparison with site-specific analysis. It was also lower for clearly defined cell subsets than those based on dim markers and for rare populations. Automated gating was able to match the performance of central manual analysis for all tested panels, exhibiting little to no bias and comparable variability. Standardized staining, data collection, and automated gating can increase power, reduce variability, and streamline analysis for immunophenotyping.

Deep learning has the potential to transform health care; however, substantial expertise is required to train such models. We sought to evaluate the utility of automated deep learning software to develop medical image diagnostic classifiers by health-care professionals with no coding-and no deep learning-expertise. We used five publicly available open-source datasets: retinal fundus images (MESSIDOR); optical coherence tomography (OCT) images (Guangzhou Medical University and Shiley Eye Institute, version 3); images of skin lesions (Human Against Machine [HAM] 10000), and both paediatric and adult chest x-ray (CXR) images (Guangzhou Medical University and Shiley Eye Institute, version 3 and the National Institute of Health [NIH] dataset, respectively) to separately feed into a neural architecture search framework, hosted through Google Cloud AutoML, that automatically developed a deep learning architecture to classify common diseases. Sensitivity (recall), specificity, and positive predictive value (precision) were used to evaluate the diagnostic properties of the models. The discriminative performance was assessed using the area under the precision recall curve (AUPRC). In the case of the deep learning model developed on a subset of the HAM10000 dataset, we did external validation using the Edinburgh Dermofit Library dataset. Diagnostic properties and discriminative performance from internal validations were high in the binary classification tasks (sensitivity 73·3-97·0%; specificity 67-100%; AUPRC 0·87-1·00). In the multiple classification tasks, the diagnostic properties ranged from 38% to 100% for sensitivity and from 67% to 100% for specificity. The discriminative performance in terms of AUPRC ranged from 0·57 to 1·00 in the five automated deep learning models. In an external validation using the Edinburgh Dermofit Library dataset, the automated deep learning model showed an AUPRC of 0·47, with a sensitivity of 49% and a positive predictive value of 52%. All models, except the automated deep learning model trained on the multilabel classification task of the NIH CXR14 dataset, showed comparable discriminative performance and diagnostic properties to state-of-the-art performing deep learning algorithms. The performance in the external validation study was low. The quality of the open-access datasets (including insufficient information about patient flow and demographics) and the absence of measurement for precision, such as confidence intervals, constituted the major limitations of this study. The availability of automated deep learning platforms provide an opportunity for the medical community to enhance their understanding in model development and evaluation. Although the derivation of classification models without requiring a deep understanding of the mathematical, statistical, and programming principles is attractive, comparable performance to expertly designed models is limited to more elementary classification tasks. Furthermore, care should be placed in adhering to ethical principles when using these automated models to avoid discrimination and causing harm. Future studies should compare several application programming interfaces on thoroughly curated datasets. National Institute for Health Research and Moorfields Eye Charity.

In a large cohort of molecularly characterized inherited retinal disease (IRD) families, we investigated proportions with disease attributable to causative variants in each gene. Retrospective study of electronic patient records. Patients and relatives managed in the Genetics Service of Moorfields Eye Hospital in whom a molecular diagnosis had been identified. Genetic screening used a combination of single-gene testing, gene panel testing, whole exome sequencing, and more recently, whole genome sequencing. For this study, genes listed in the Retinal Information Network online resource (https://sph.uth.edu/retnet/) were included. Transcript length was extracted for each gene (Ensembl, release 94). We calculated proportions of families with IRD attributable to variants in each gene in the entire cohort, a cohort younger than 18 years, and a current cohort (at least 1 patient encounter between January 1, 2017, and August 2, 2019). Additionally, we explored correlation between numbers of families and gene transcript length. We identified 3195 families with a molecular diagnosis (variants in 135 genes), including 4236 affected individuals. The pediatric cohort comprised 452 individuals from 411 families (66 genes). The current cohort comprised 2614 families (131 genes; 3130 affected individuals). The 20 most frequently implicated genes overall (with prevalence rates per families) were as follows: ABCA4 (20.8%), USH2A (9.1%), RPGR (5.1%), PRPH2 (4.6%), BEST1 (3.9%), RS1 (3.5%), RP1 (3.3%), RHO (3.3%), CHM (2.7%), CRB1 (2.1%), PRPF31 (1.8%), MY07A (1.7%), OPA1 (1.6%), CNGB3 (1.4%), RPE65 (1.2%), EYS (1.2%), GUCY2D (1.2%), PROM1 (1.2%), CNGA3 (1.1%), and RDH12 (1.1%). These accounted for 71.8% of all molecularly diagnosed families. Spearman coefficients for correlation between numbers of families and transcript length were 0.20 (P = 0.025) overall and 0.27 (P = 0.017), -0.17 (P = 0.46), and 0.71 (P = 0.047) for genes in which variants exclusively cause recessive, dominant, or X-linked disease, respectively. Our findings help to quantify the burden of IRD attributable to each gene. More than 70% of families showed pathogenic variants in 1 of 20 genes. Transcript length (relevant to gene delivery strategies) correlated significantly with numbers of affected families (but not for dominant disease).

A number of large technology companies have created code-free cloud-based platforms that allow researchers and clinicians without coding experience to create deep learning algorithms. In this study, we comprehensively analyse the performance and featureset of six platforms, using four representative cross-sectional and en-face medical imaging datasets to create image classification models. The mean (s.d.) F1 scores across platforms for all model–dataset pairs were as follows: Amazon, 93.9 (5.4); Apple, 72.0 (13.6); Clarifai, 74.2 (7.1); Google, 92.0 (5.4); MedicMind, 90.7 (9.6); Microsoft, 88.6 (5.3). The platforms demonstrated uniformly higher classification performance with the optical coherence tomography modality. Potential use cases given proper validation include research dataset curation, mobile ‘edge models’ for regions without internet access, and baseline models against which to compare and iterate bespoke deep learning approaches.