Performance of a Convolutional Neural Network and Explainability Technique for 12-Lead Electrocardiogram Interpretation.

Importance: Millions of clinicians rely daily on automated preliminary electrocardiogram (ECG) interpretation. Critical comparisons of machine learning-based automated analysis against clinically accepted standards of care are lacking. Objective: To use readily available 12-lead ECG data to train and apply an explainability technique to a convolutional neural network (CNN) that achieves high performance against clinical standards of care. Design, Setting, and Participants: This cross-sectional study was conducted using data from January 1, 2003, to December 31, 2018. Data were obtained in a commonly available 12-lead ECG format from a single-center tertiary care institution. All patients aged 18 years or older who received ECGs at the University of California, San Francisco, were included, yielding a total of 365 009 patients. Data were analyzed from January 1, 2019, to March 2, 2021. Exposures: A CNN was trained to predict the presence of 38 diagnostic classes in 5 categories from 12-lead ECG data. A CNN explainability technique called LIME (Linear Interpretable Model-Agnostic Explanations) was used to visualize ECG segments contributing to CNN diagnoses. Main Outcomes and Measures: Area under the receiver operating characteristic curve (AUC), sensitivity, and specificity were calculated for the CNN in the holdout test data set against cardiologist clinical diagnoses. For a second validation, 3 electrophysiologists provided consensus committee diagnoses against which the CNN, cardiologist clinical diagnosis, and MUSE (GE Healthcare) automated analysis performance was compared using the F1 score; AUC, sensitivity, and specificity were also calculated for the CNN against the consensus committee. Results: A total of 992 748 ECGs from 365 009 adult patients (mean [SD] age, 56.2 [17.6] years; 183 600 women [50.3%]; and 175 277 White patients [48.0%]) were included in the analysis. In 91 440 test data set ECGs, the CNN demonstrated an AUC of at least 0.960 for 32 of 38 classes (84.2%). Against the consensus committee diagnoses, the CNN had higher frequency-weighted mean F1 scores than both cardiologists and MUSE in all 5 categories (CNN frequency-weighted F1 score for rhythm, 0.812; conduction, 0.729; chamber diagnosis, 0.598; infarct, 0.674; and other diagnosis, 0.875). For 32 of 38 classes (84.2%), the CNN had AUCs of at least 0.910 and demonstrated comparable F1 scores and higher sensitivity than cardiologists, except for atrial fibrillation (CNN F1 score, 0.847 vs cardiologist F1 score, 0.881), junctional rhythm (0.526 vs 0.727), premature ventricular complex (0.786 vs 0.800), and Wolff-Parkinson-White (0.800 vs 0.842). Compared with MUSE, the CNN had higher F1 scores for all classes except supraventricular tachycardia (CNN F1 score, 0.696 vs MUSE F1 score, 0.714). The LIME technique highlighted physiologically relevant ECG segments. Conclusions and Relevance: The results of this cross-sectional study suggest that readily available ECG data can be used to train a CNN algorithm to achieve comparable performance to clinical cardiologists and exceed the performance of MUSE automated analysis for most diagnoses, with some exceptions. The LIME explainability technique applied to CNNs highlights physiologically relevant ECG segments that contribute to the CNN's diagnoses.

Correction: Correction: A Novel Rhabdovirus Associated with Acute Hemorrhagic Fever in Central Africa.

[This corrects the article DOI: 10.1371/journal.ppat.1002924.].

Multi-gene panel testing for hereditary cancer predisposition in unsolved high-risk breast and ovarian cancer patients.

PURPOSE: Many women with an elevated risk of hereditary breast and ovarian cancer have previously tested negative for pathogenic mutations in BRCA1 and BRCA2. Among them, a subset has hereditary susceptibility to cancer and requires further testing. We sought to identify specific groups who remain at high risk and evaluate whether they should be offered multi-gene panel testing. METHODS: We tested 300 women on a multi-gene panel who were previously enrolled in a long-term study at UCSF. As part of their long-term care, all previously tested negative for mutations in BRCA1 and BRCA2 either by limited or comprehensive sequencing. Additionally, they met one of the following criteria: (i) personal history of bilateral breast cancer, (ii) personal history of breast cancer and a first or second degree relative with ovarian cancer, and (iii) personal history of ovarian, fallopian tube, or peritoneal carcinoma. RESULTS: Across the three groups, 26 women (9%) had a total of 28 pathogenic mutations associated with hereditary cancer susceptibility, and 23 women (8%) had mutations in genes other than BRCA1 and BRCA2. Ashkenazi Jewish and Hispanic women had elevated pathogenic mutation rates. In addition, two women harbored pathogenic mutations in more than one hereditary predisposition gene. CONCLUSIONS: Among women at high risk of breast and ovarian cancer who have previously tested negative for pathogenic BRCA1 and BRCA2 mutations, we identified three groups of women who should be considered for subsequent multi-gene panel testing. The identification of women with multiple pathogenic mutations has important implications for family testing.

A cloud-compatible bioinformatics pipeline for ultrarapid pathogen identification from next-generation sequencing of clinical samples.

Unbiased next-generation sequencing (NGS) approaches enable comprehensive pathogen detection in the clinical microbiology laboratory and have numerous applications for public health surveillance, outbreak investigation, and the diagnosis of infectious diseases. However, practical deployment of the technology is hindered by the bioinformatics challenge of analyzing results accurately and in a clinically relevant timeframe. Here we describe SURPI ("sequence-based ultrarapid pathogen identification"), a computational pipeline for pathogen identification from complex metagenomic NGS data generated from clinical samples, and demonstrate use of the pipeline in the analysis of 237 clinical samples comprising more than 1.1 billion sequences. Deployable on both cloud-based and standalone servers, SURPI leverages two state-of-the-art aligners for accelerated analyses, SNAP and RAPSearch, which are as accurate as existing bioinformatics tools but orders of magnitude faster in performance. In fast mode, SURPI detects viruses and bacteria by scanning data sets of 7-500 million reads in 11 min to 5 h, while in comprehensive mode, all known microorganisms are identified, followed by de novo assembly and protein homology searches for divergent viruses in 50 min to 16 h. SURPI has also directly contributed to real-time microbial diagnosis in acutely ill patients, underscoring its potential key role in the development of unbiased NGS-based clinical assays in infectious diseases that demand rapid turnaround times.

Toward an understanding of changes in diversity associated with fecal microbiome transplantation based on 16S rRNA gene deep sequencing.

Fecal microbiome transplantation by low-volume enema is an effective, safe, and inexpensive alternative to antibiotic therapy for patients with chronic relapsing Clostridium difficile infection (CDI). We explored the microbial diversity of pre- and posttransplant stool specimens from CDI patients (n = 6) using deep sequencing of the 16S rRNA gene. While interindividual variability in microbiota change occurs with fecal transplantation and vancomycin exposure, in this pilot study we note that clinical cure of CDI is associated with an increase in diversity and richness. Genus- and species-level analysis may reveal a cocktail of microorganisms or products thereof that will ultimately be used as a probiotic to treat CDI. IMPORTANCE Antibiotic-associated diarrhea (AAD) due to Clostridium difficile is a widespread phenomenon in hospitals today. Despite the use of antibiotics, up to 30% of patients are unable to clear the infection and suffer recurrent bouts of diarrheal disease. As a result, clinicians have resorted to fecal microbiome transplantation (FT). Donor stool for this type of therapy is typically obtained from a spouse or close relative and thoroughly tested for various pathogenic microorganisms prior to infusion. Anecdotal reports suggest a very high success rate of FT in patients who fail antibiotic treatment (>90%). We used deep-sequencing technology to explore the human microbial diversity in patients with Clostridium difficile infection (CDI) disease after FT. Genus- and species-level analysis revealed a cocktail of microorganisms in the Bacteroidetes and Firmicutes phyla that may ultimately be used as a probiotic to treat CDI.

A novel rhabdovirus associated with acute hemorrhagic fever in central Africa.

Deep sequencing was used to discover a novel rhabdovirus (Bas-Congo virus, or BASV) associated with a 2009 outbreak of 3 human cases of acute hemorrhagic fever in Mangala village, Democratic Republic of Congo (DRC), Africa. The cases, presenting over a 3-week period, were characterized by abrupt disease onset, high fever, mucosal hemorrhage, and, in two patients, death within 3 days. BASV was detected in an acute serum sample from the lone survivor at a concentration of 1.09 × 10(6) RNA copies/mL, and 98.2% of the genome was subsequently de novo assembled from ≈ 140 million sequence reads. Phylogenetic analysis revealed that BASV is highly divergent and shares less than 34% amino acid identity with any other rhabdovirus. High convalescent neutralizing antibody titers of >1:1000 were detected in the survivor and an asymptomatic nurse directly caring for him, both of whom were health care workers, suggesting the potential for human-to-human transmission of BASV. The natural animal reservoir host or arthropod vector and precise mode of transmission for the virus remain unclear. BASV is an emerging human pathogen associated with acute hemorrhagic fever in Africa.

A metagenomic analysis of pandemic influenza A (2009 H1N1) infection in patients from North America.

Although metagenomics has been previously employed for pathogen discovery, its cost and complexity have prevented its use as a practical front-line diagnostic for unknown infectious diseases. Here we demonstrate the utility of two metagenomics-based strategies, a pan-viral microarray (Virochip) and deep sequencing, for the identification and characterization of 2009 pandemic H1N1 influenza A virus. Using nasopharyngeal swabs collected during the earliest stages of the pandemic in Mexico, Canada, and the United States (n = 17), the Virochip was able to detect a novel virus most closely related to swine influenza viruses without a priori information. Deep sequencing yielded reads corresponding to 2009 H1N1 influenza in each sample (percentage of aligned sequences corresponding to 2009 H1N1 ranging from 0.0011% to 10.9%), with up to 97% coverage of the influenza genome in one sample. Detection of 2009 H1N1 by deep sequencing was possible even at titers near the limits of detection for specific RT-PCR, and the percentage of sequence reads was linearly correlated with virus titer. Deep sequencing also provided insights into the upper respiratory microbiota and host gene expression in response to 2009 H1N1 infection. An unbiased analysis combining sequence data from all 17 outbreak samples revealed that 90% of the 2009 H1N1 genome could be assembled de novo without the use of any reference sequence, including assembly of several near full-length genomic segments. These results indicate that a streamlined metagenomics detection strategy can potentially replace the multiple conventional diagnostic tests required to investigate an outbreak of a novel pathogen, and provide a blueprint for comprehensive diagnosis of unexplained acute illnesses or outbreaks in clinical and public health settings.

Concerted potent humoral immune responses to autoantigens are associated with tumor destruction and favorable clinical outcomes without autoimmunity.

PURPOSE: The therapeutic importance of immune responses against single versus multiple antigens is poorly understood. There also remains insufficient understanding whether responses to one subset of antigens are more significant than another. Autoantibodies are frequent in cancer patients. They can pose no biological significance or lead to debilitating paraneoplastic syndromes. Autoreactivity has been associated with clinical benefits, but the magnitude necessary for meaningful results is unknown. Autologous tumor cells engineered to secrete granulocyte macrophage colony-stimulating factor generate immune infiltrates in preexisting metastases with associated tumor destruction. We sought to identify targets of responses from this vaccination strategy. EXPERIMENTAL DESIGN: Postvaccination sera used in screening a cDNA expression library prepared from a densely infiltrated metastasis of a long-term surviving melanoma patient identified several autoantigens. Additional autoantigens were identified through similar screenings in non-small cell lung cancer and murine models, and proteins implicated in cancer propagation. ELISAs for several targets were established using recombinant proteins, whereas others were evaluated by petit serologies. RESULTS: Eleven gene products were identified through serologic screening from two patients showing highly favorable clinical outcomes. A subset of antigens revealed significant changes in antibody titers compared with weak responses to other proteins. Time course analyses showed coordinated enhanced titers against several targets as a function of vaccination in responding patients. CONCLUSIONS: This study shows the range of biologically significant antigens resulting from a whole-cell vaccine. Targets include autoantigens that are components of cell cycle regulation. Potent antibody responses against multiple autoantigens are associated with effective tumor destruction without clinical autoimmunity.

The Plasmodium protein network diverges from those of other eukaryotes.

Plasmodium falciparum is the pathogen responsible for over 90% of human deaths from malaria. Therefore, it has been the focus of a considerable research initiative, involving the complete DNA sequencing of the genome, large-scale expression analyses, and protein characterization of its life-cycle stages. The Plasmodium genome sequence is relatively distant from those of most other eukaryotes, with more than 60% of the 5,334 encoded proteins lacking any notable sequence similarity to other organisms. To systematically elucidate functional relationships among these proteins, a large two-hybrid study has recently mapped a network of 2,846 interactions involving 1,312 proteins within Plasmodium. This network adds to a growing collection of available interaction maps for a number of different organisms, and raises questions about whether the divergence of Plasmodium at the sequence level is reflected in the configuration of its protein network. Here we examine the degree of conservation between the Plasmodium protein network and those of model organisms. Although we find 29 highly connected protein complexes specific to the network of the pathogen, we find very little conservation with complexes observed in other organisms (three in yeast, none in the others). Overall, the patterns of protein interaction in Plasmodium, like its genome sequence, set it apart from other species.

Conserved patterns of protein interaction in multiple species.

To elucidate cellular machinery on a global scale, we performed a multiple comparison of the recently available protein-protein interaction networks of Caenorhabditis elegans, Drosophila melanogaster, and Saccharomyces cerevisiae. This comparison integrated protein interaction and sequence information to reveal 71 network regions that were conserved across all three species and many exclusive to the metazoans. We used this conservation, and found statistically significant support for 4,645 previously undescribed protein functions and 2,609 previously undescribed protein interactions. We tested 60 interaction predictions for yeast by two-hybrid analysis, confirming approximately half of these. Significantly, many of the predicted functions and interactions would not have been identified from sequence similarity alone, demonstrating that network comparisons provide essential biological information beyond what is gleaned from the genome.

Conserved pathways within bacteria and yeast as revealed by global protein network alignment.

We implement a strategy for aligning two protein-protein interaction networks that combines interaction topology and protein sequence similarity to identify conserved interaction pathways and complexes. Using this approach we show that the protein-protein interaction networks of two distantly related species, Saccharomyces cerevisiae and Helicobacter pylori, harbor a large complement of evolutionarily conserved pathways, and that a large number of pathways appears to have duplicated and specialized within yeast. Analysis of these findings reveals many well characterized interaction pathways as well as many unanticipated pathways, the significance of which is reinforced by their presence in the networks of both species.