Human Reading Research Articles

Application of Supervised Machine Learning to Recognize Competent Level and Mixed Antinuclear Antibody Patterns Based on ICAP International Consensus.

Background: Antinuclear antibody pattern recognition is vital for autoimmune disease diagnosis but labor-intensive for manual interpretation. To develop an automated pattern recognition system, we established machine learning models based on the International Consensus on Antinuclear Antibody Patterns (ICAP) at a competent level, mixed patterns recognition, and evaluated their consistency with human reading. Methods: 51,694 human epithelial cells (HEp-2) cell images with patterns assigned by experienced medical technologists collected in a medical center were used to train six machine learning algorithms and were compared by their performance. Next, we choose the best performing model to test the consistency with five experienced readers and two beginners. Results: The mean F1 score in each classification of the best performing model was 0.86 evaluated by Testing Data 1. For the inter-observer agreement test on Testing Data 2, the average agreement was 0.849 (κ) among five experienced readers, 0.844 between the best performing model and experienced readers, 0.528 between experienced readers and beginners. The results indicate that the proposed model outperformed beginners and achieved an excellent agreement with experienced readers. Conclusions: This study demonstrated that the developed model could reach an excellent agreement with experienced human readers using machine learning methods.

Learning with a Beat: A Study on the Implementation of Students’ Music Interests in High School Humanities Curriculum

From the first-ever Pulitzer Prize for a hip-hop album awarded to rapper Kendrick Lamar in 2014, to the development of entire websites dedicated to popular music analyses, music has become a form of accessible humanities-like education, as people listen to the deeper meaning within a song's beat. Contrarily, the traditional humanities education in today’s high schools has hardly seen change, if any at all, with student apathy rapidly rising. Thus, this research paper asks, "Can the music interests of high-achieving high school students be implemented in humanities-based curricula to combat apathy in learning? If so, how?" The study was conducted in a three-part mixed methods approach. First, an action research survey collected data on the favorite "thought-provoking" songs of high-achieving high school students from two high schools. Next, content analyses on the three songs most frequently mentioned by students were conducted based on humanities-derived categories– theme, rhetoric, and context. Finally, the content analyses were compiled into a hermeneutics-based survey for head teachers in the humanities (reading, English Language Arts, and social studies) departments at the schools. This study supports the general conclusion that the music interests of high-achieving students can be implemented into humanities-based curricula, in multiple ways, to combat learning apathy. The results of this study give a proven basis for reducing apathy in humanities-based learning with music, as the vast majority of students and teachers surveyed advocate for this implementation. Teachers’ reported interpretations provide a framework for new curricula that could improve students’ learning appeal and engagement.

Characterization of Cell Free Plasma Methyl-DNA From Xenografted Tumors to Guide the Selection of Diagnostic Markers for Early-Stage Cancers

Circulating cell-free methyl-DNA (mcfDNA) contains promising cancer markers but its low abundance and possibly diverse origin pose challenges toward the accurate diagnosis of early stage cancers. By whole-genome bisulfite sequencing (WGBS) of cell-free DNA (cfDNA) from about 0.5 mL plasma of mice xenografted with human tumors, we obtained and aligned the reads to the human genome, filtered out the mouse and carrier bacterial sequences, and confirmed the tumor origin of methyl-cfDNA (mctDNA) by methylation-sensitive restriction enzyme digestion prior to species-specific PCR. We estimated that human tumor-specific reads (ctDNA) or mctDNA comprised about 0.29 or 0.01%, respectively of the xenograft mouse cfDNA, and about 0.029 or 0.001% of the cfDNA of human early stage cancer patients. Similar WGBS of early stage (0-II, node- and metastasis-free) breast, lung or colorectal cancer samples identified hundreds of specific DMRs (differentially methylated regions) compared to healthy controls. Their association with tumourigenesis was supported by stage-dependent methylation, tumor suppressor or oncogene clusters, and genes also identified in the xenograft samples. Using 20 three-cancer-common and 17 colorectal cancer-specific DMRs in combination (top 0.0018% of the WGBS methylation clusters) was sufficient to distinguish the stage I colorectal cancers from breast and lung cancers and healthy controls. Our data thus confirmed the tumor origin of mctDNA by sequence specificity, and provide a selection threshold for authentic tumor mctDNA markers toward precise diagnosis of early stage cancers solely by top DMRs in combination.

Galvanic skin response to recognizing human behaviour

Human beings' health is affected by physical ventures and emotional states endured by regular activities, which frequently develop attitudes and substantially affect health outcomes. Human Emotions play a vital role in deciding perception, cognition, memory, attention, reasoning, and decision-making. Several approaches have been used for automatically recognizing users' sentiment through images, speech, text, video, and physiological signals. Truthful detection of human emotions and personality behaviours can be advantageous for many situations, like interviews, group discussions, polygraphs, depressed persons, paralytic patients, blind people, shooters, etc. So, there is a need for an emotion recognizer. Though researchers tried several methods for emotion recognition, the accuracy of detection is always a question. The main aim is to develop a precise classification model for better accuracy of the emotion recognition system. Therefore, an emotion detector using GSR (Grove – GSR Sensor V1.2) sensor is proposed in the current research work. Twenty pupil groups were subjected to under observation for six different human activities, viz., happy, relax, stress, pain, reading, and math calculation. This research work was carried out in the lab of machine vision (CEERI) Central Electronics Engineering Research Institute Pilani, Jhunjhunu India. Moving average window method was used for data pre-processing. Supervised machine learning models viz., k-nearest neighbours (KNN), support vector machine (SVM), and decision tree (DT) was used for emotion classification. The decision tree model gives the best results with an average accuracy of 97.61%. Pain activity is most correctly recognized with greater than 99% accuracy.

Distinct genomic profile in h. pylori-associated gastric cancer.

Gastric cancer is one of the most common and deadly cancer types. Currently, four subtypes have been identified with unique molecular alterations: Epstein–Barr virus (EBV)‐positive, microsatellite instability (MSI), chromosomal instability (CIN), and genomic stable (GS) tumors. Notably, many gastric tumors are associated with the bacterium Helicobacter pylori but the genomic landscape of this subgroup of tumors remains largely unknown. Targeted sequencing covering 425 genes was performed retrospectively on 1703 gastric tumor tissues and matched normal blood samples. Nonsynonymous mutations, copy‐number variation (CNV), and MSI status were called from human DNA reads; nonhuman DNA reads were mapped to NCBI microbial reference genome using Kraken and significant species were identified. Overall, 37 (2.76%) from a total of 1703 samples were EBV‐positive, 200 (11.74%) samples were H. pylori‐positive, and 10 samples were positive for both. Among the rest, 59 (3.46%) samples were MSI, 380 (22.31%) were CIN, and 1017 (59.72%) were GS. Most of the 200 H. pylori‐positive samples tend to be genome stable (85.5%, p < 0.001) and microsatellite stable (95%, p = 0.04). Compared to 1017 GS tumors, mutations in AKT3, EPAS1, MLH1, and BKT and amplifications of NFE2L2, TERC, MCL1, and TOP1 were significantly enriched in H. pylori‐positive tumors. And compared to EBV‐positive tumors, mutations in PIK3CA, ARID1A, and PTEN were significantly depleted in H. pylori‐positive subtype while TP53 mutations were enriched. This study characterized the unique genomic landscape of H. pylori‐positive gastric tumors using targeted panel sequencing. The successful identification of DNA reads from infectious agents in tumor samples indicates that deep sequencing is a promising way to uncover characteristics of microbial environment in tumors.

Abstract PO-049: Pseudo-alignment resolution in discriminating mouse reads

Abstract Patient-Derived Xenografts (PDX) are the best preclinical models to study cancer biology preclinical drug screening. PDXs are generated by implanting a piece of fresh cancer tissue directly into an immunodeficient mouse to grow and imitate a human tumor. Previous studies have shown that PDXs are able to accurately recapitulate the (epi-)genomic characteristics of the original tumor. Yet the inclusion of the mouse cells in tumor biopsies can introduce unwanted noise and bias in high throughput (epi-)genomics analyses and interpretation. Thus, filtering the mouse reads is one of the first and important steps in analyzing high throughput PDX sequencing data. There are two main approaches to account for mouse reads in analyzing bulk RNAseq data generated from PDXs. In the first approach, an external tool is used to classify the raw or aligned reads into a human, mouse, or ambiguous reads. Examples include XenoFilter, BBsplit, disambiguate, and Xenome. In the second approach, a mixed human-mouse reference genome is used when aligning the raw reads. Multiple studies have shown that the mixed reference approach produces better or similar results compared to when an external tool is used for mouse read discrimination. However, these comparisons have employed exact alignment methods when using a mixed reference approach. Recently, the significant deluge of sequencing data has brought attention to pseudo-alignment based methods, specifically Kallisto and Salmon. By taking advantage of pseudo-alignment to a reference transcriptome, these tools are capable of analyzing millions of reads in minutes while demonstrating near-optimal quantification performance. Yet their performance in analyzing PDX RNAseq data while using a mixed reference is unclear. Moreover, among the external methods, only Xenome has a pseudo-alignment strategy for read classification and the rest of the methods use exact alignment making them labor-intensive and intractable for large datasets. In this study, we compare two main mouse read discrimination pipelines when using Kallisto and Salmon. By sampling from two human and mouse lung RNAseq datasets, we have simulated five pseudo-PDX RNAseq datasets with different contamination ratios. By comparing outputs of different discriminating strategies to the ground truth, we show that Kallisto and Salmon when used with mixed human-mouse transcriptome produce results that are more correlated with the ground truth. Moreover, by differential gene expression analyses we show that fewer genes are significantly affected when using the mixed reference approach. Citation Format: Soheil Jahangiri-Tazehkand, Farnoosh Agha-Babazadeh, Benjamin Haibe-Kains. Pseudo-alignment resolution in discriminating mouse reads [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-049.

Tumor hepatitis B virus RNA identifies a clinically and molecularly distinct subset of hepatocellular carcinoma.

Hepatitis B virus (HBV) infection contributes to hepatocellular carcinoma (HCC) initiation and is associated with worse outcomes. Many prior studies of HBV-related HCC have not accounted for potential heterogeneity among HBV-related tumors by assessing whether HBV activity is present in tumor tissue. Here, we measured tumor HBV RNA, a proxy for viral activity, and investigated the association between HBV RNA status and several clinicogenomic characteristics. We obtained clinical, mutation, RNA-Seq and survival data for 439 HCC tumors from The Cancer Genome Atlas and International Cancer Genome Consortium. Tumors were classified as HBV RNA positive if they harbored >1 HBV RNA read per million human reads. We investigated the association between HBV RNA status and nonsynonymous somatic mutations, gene set expression, homologous recombination deficiency (HRD) score and mutation-specific survival. HBV RNA positive status was associated with higher nonsynonymous mutation rates of multiple genes, including TP53 and CDKN2A, while HBV RNA negative status was associated with higher nonsynonymous BAP1 mutation rate. HBV RNA positive status was also associated with increased transcription of genes involved in multiple DNA damage repair pathways, genes upregulated by MYC and mTORC1, and genes overexpressed in several HCC subclasses associated with a proliferative phenotype. Further, HBV RNA positive status was associated with increased three-biomarker HRD score (22.2 for HBV RNA+ vs. 16.0 for HBV RNA-). Finally, HBV RNA status was associated with multiple mutation-specific survival differences, including decreased survival for HBV RNA positive patients with nonsynonymous KEAP1 mutations compared to those without (hazard ratio 4.26). HCC tumors harboring genomic evidence of HBV activity therefore constitute a distinct HCC subset characterized by specific differences in nonsynonymous mutations, gene set expression, three-biomarker HRD score and mutation-specific survival.

DeepAMO: a multi-slice, multi-view anthropomorphic model observer for visual detection tasks performed on volume images.

Purpose: We propose a deep learning-based anthropomorphic model observer (DeepAMO) for image quality evaluation of multi-orientation, multi-slice image sets with respect to a clinically realistic 3D defect detection task. Approach: The DeepAMO is developed based on a hypothetical model of the decision process of a human reader performing a detection task using a 3D volume. The DeepAMO is comprised of three sequential stages: defect segmentation, defect confirmation (DC), and rating value inference. The input to the DeepAMO is a composite image, typical of that used to view 3D volumes in clinical practice. The output is a rating value designed to reproduce a human observer's defect detection performance. In stages 2 and 3, we propose: (1)a projection-based DC block that confirms defect presence in two 2D orthogonal orientations and (2)a calibration method that "learns" the mapping from the features of stage 2 to the distribution of observer ratings from the human observer rating data (thus modeling inter- or intraobserver variability) using a mixture density network. We implemented and evaluated the DeepAMO in the context of -DMSA SPECT imaging. A human observer study was conducted, with two medical imaging physics graduate students serving as observers. A -fold cross-validation experiment was conducted to test the statistical equivalence in defect detection performance between the DeepAMO and the human observer. We also compared the performance of the DeepAMO to an unoptimized implementation of a scanning linear discriminant observer (SLDO). Results: The results show that the DeepAMO's and human observer's performances on unseen images were statistically equivalent with a margin of difference ( ) of 0.0426 at , using 288 training images. A limited implementation of an SLDO had a substantially higher AUC (0.99) compared to the DeepAMO and human observer. Conclusion: The results show that the DeepAMO has the potential to reproduce the absolute performance, and not just the relative ranking of human observers on a clinically realistic defect detection task, and that building conceptual components of the human reading process into deep learning-based models can allow training of these models in settings where limited training images are available.

MBG: Minimizer-based sparse de Bruijn Graph construction.

MotivationDe Bruijn graphs can be constructed from short reads efficiently and have been used for many purposes. Traditionally, long-read sequencing technologies have had too high error rates for de Bruijn graph-based methods. Recently, HiFi reads have provided a combination of long-read length and low error rate, which enables de Bruijn graphs to be used with HiFi reads.ResultsWe have implemented MBG, a tool for building sparse de Bruijn graphs from HiFi reads. MBG outperforms existing tools for building dense de Bruijn graphs and can build a graph of 50× coverage whole human genome HiFi reads in four hours on a single core. MBG also assembles the bacterial E.coli genome into a single contig in 8 s.Availability and implementationPackage manager: https://anaconda.org/bioconda/mbg and source code: https://github.com/maickrau/MBG.Supplementary informationSupplementary data are available at Bioinformatics online.

Diabetic Retinopathy Screening Using Artificial Intelligence and Handheld Smartphone-Based Retinal Camera.

Portable retinal cameras and deep learning (DL) algorithms are novel tools adopted by diabetic retinopathy (DR) screening programs. Our objective is to evaluate the diagnostic accuracy of a DL algorithm and the performance of portable handheld retinal cameras in the detection of DR in a large and heterogenous type 2 diabetes population in a real-world, high burden setting. Participants underwent fundus photographs of both eyes with a portable retinal camera (Phelcom Eyer). Classification of DR was performed by human reading and a DL algorithm (PhelcomNet), consisting of a convolutional neural network trained on a dataset of fundus images captured exclusively with the portable device; both methods were compared. We calculated the area under the curve (AUC), sensitivity, and specificity for more than mild DR. A total of 824 individuals with type 2 diabetes were enrolled at Itabuna Diabetes Campaign, a subset of 679 (82.4%) of whom could be fully assessed. The algorithm sensitivity/specificity was 97.8 % (95% CI 96.7-98.9)/61.4 % (95% CI 57.7-65.1); AUC was 0·89. All false negative cases were classified as moderate non-proliferative diabetic retinopathy (NPDR) by human grading. The DL algorithm reached a good diagnostic accuracy for more than mild DR in a real-world, high burden setting. The performance of the handheld portable retinal camera was adequate, with over 80% of individuals presenting with images of sufficient quality. Portable devices and artificial intelligence tools may increase coverage of DR screening programs.

Incremental Text-to-Speech Synthesis Using Pseudo Lookahead With Large Pretrained Language Model

This letter presents an incremental text-to-speech (TTS) method that performs synthesis in small linguistic units while maintaining the naturalness of output speech. Incremental TTS is generally subject to a trade-off between latency and synthetic speech quality. It is challenging to produce high-quality speech with a low-latency setup that does not make much use of an unobserved future sentence (hereafter, "lookahead"). To resolve this issue, we propose an incremental TTS method that uses a pseudo lookahead generated with a language model to take the future contextual information into account without increasing latency. Our method can be regarded as imitating a human's incremental reading and uses pretrained GPT2, which accounts for the large-scale linguistic knowledge, for the lookahead generation. Evaluation results show that our method 1) achieves higher speech quality than the method taking only observed information into account and 2) achieves a speech quality equivalent to waiting for the future context observation.

An automatic machine learning approach for ischemic stroke onset time identification based on DWI and FLAIR imaging

Current thrombolysis for acute ischemic stroke (AIS) treatment strictly relies on the time since stroke (TSS) less than 4.5 h. However, some patients are excluded from thrombolytic treatment because of the unknown TSS. The diffusion-weighted imaging (DWI) and fluid-attenuated inversion recovery (FLAIR) mismatch can simply identify TSS since lesion intensities are not identical at different onset time. In this paper, we propose an automatic machine learning method to classify the TSS less than or more than 4.5 h. First, we develop a cross-modal convolutional neural network to accurately segment the stroke lesions from DWI and FLAIR images. Second, the features are extracted from DWI and FLAIR according to the segmentation regions of interest (ROI). Finally, the features are fed to machine learning models to identify TSS. In DWI and FLAIR ROI segmentation, the networks obtain high Dice coefficients with 0.803 and 0.647. The classification test results show that our model achieves an accuracy of 0.805, with a sensitivity of 0.769 and a specificity of 0.840. Our approach outperforms human reading DWI-FLAIR mismatch model, illustrating the potential for automatic and fast TSS identification.

Analysis and simulation of handwritten recognition system

Handwriting recognition is the automatic transcription of handwriting, where only the image of the handwriting is available. Manual matching shall be used by banks for the authentication of checks and signatures. In forensics, handwriting corresponding algorithms can help handwriting analysts to predict the author with greater precision. This handwriting needs to be scanned to the computer for the handwriting recognition system to access it and analyse it consequently. A variety of handwriting applications, including transcription papers, mail routing, and processor forms, checks, and faxes, may be envisaged. Several applications are also possible. The extensive research effoncentrated on the field of character recognition (CR), due both to its possible applications and to the difficulties involved in simulating human reading. The new Offline Handwritten Recognition (OHR) is designed for both Urdu and English. It mainly focuses on the removal of noises in word, character segmentation methods with higher recognition rate. The images which are scanned may contain noises and image de-noising steps consist of binarization, noise elimination, and size normalization. Words and character segmentation are performed by using Particle Swarm Optimization (PSO) algorithm. Then those segmented samples are used for the next step which is feature extraction. Finally, word recognition is performed by using the deep neural network classifier.

Machine Learning for Determining Lateral Flow Device Results in Asymptomatic Population: A Diagnostic Accuracy Study

Introduction: Rapid antigen tests, in the form of lateral flow devices (LFD) allow testing of a large population for SARS-CoV-2 rapidly either on an asymptomatic or symptomatic basis and have been demonstrated to reduce transmission when used correctly. Compared to real time quantitative PCR (RT-qPCR), LFD have a different sensitivity range to viral load, and can be used to self-test, which can present challenges for lay users to interpret especially at low viral loads. In order to reduce the variability seen in device interpretation, we designed and tested an AI algorithm based on machine learning. Methods: A machine learning (ML) algorithm was trained on a combination of artificially hybridised LFDs from a range of test conditions and viral dilutions and real-world LFD data linked to RT-qPCR results. Two sub-studies were carried out. In the first, participants were recruited from asymptomatic test sites (ATS) in the community and in the second from health care workers undertaking self-testing using LFDs at home. A panel of trained clinicians were used to resolve discrepancies between the algorithm and the ATS/self-report result. Standard estimates of diagnostic precision were applied, comparing ATS and self-report results to the diagnostic algorithm, with and without resolution of discrepancies. Results: In total, 115,316 images were returned by the ATS and self-test users (59,450 ATS, 50,999 self-test). In the ATS sub study, use of the AI algorithm increased sensitivity compared to a human reader from 92.08% to 97.6% (95% CI 93.20-99.51%) and specificity from 99.85% to 99.99% (95% CI -99.90%), correctly identifying another 30 positive images compared to the submitted result. In the self-read sub study, sensitivity increased from 16.00% to 100% (95% CI 90.2-100%), predominantly due to detection of faint bands representing low viral loads, and specificity from 99.15% to 99.40% (95% CI 99.33-99.47%), with an additional 32 positive images correctly identified compared to the submitted result. Conclusion: An ML-based classifier of LFD results outperforms human reads in asymptomatic testing sites and self-reading. Clinical Trial Registration Details: This study is registered with the ISRCTN registry (reference number 39894, https://www.isrctn.com/ISRCTN30075312. Funding Information: The study was funded by NHS Track and Trace. ADB is funded by an CRUK Advanced Clinician Scientist Award (ref C31641/A23923). Declaration of Interests: There are no competing interests. Ethics Approval Statement: Ethical approval was sought from the Public Health England Research Ethics committee, who confirmed that the study was exempt from ethical approval, as the study was purely observational with the AI result not being returned to the user.

390. Non-invasive Detection of Co-infections in Hospitalized Patients with COVID-19 using the Karius Test, A Plasma-based Next-Generation Sequencing Test for Microbial Cell-free DNA

BackgroundPatients hospitalized with SARS-CoV2 infections (Covid-19) are frequently febrile and can become critically ill quickly often leading to intervention with antimicrobial therapy. An etiologic diagnosis of superinfections can be difficult to obtain through the usual invasive procedures because of patient instability and the desire to avoid them because they may not be tolerated by the patient. Providers may also be hesitant to embark on such interventions in order to avoid healthcare personnel (HCP) exposure to aerosols.MethodsKarius Test (KT) results are presented from 30 patients who presented with Covid-19. The KT is a CLIA certified/CAP-accredited next-generation sequencing (NGS) plasma test that detects pathogen cell free DNA (cfDNA). After cfDNA is extracted and NGS performed, human reads are removed and remaining sequences are aligned to a curated database of > 1400 organisms. Organisms present above a statistical threshold are reported.ResultsThe KT detected pathogens in the majority of patients (n=20) with COVID-19. The most common infections were herpesviruses in 60% of patients. The most common bacterial pathogen was E. coli seen in 25% of patients. 15 out of 20 patients had more than one pathogen detected. 15% of patients had fungal pathogens, including one detection of Lichtheimia ramosa, in an immunocompromised patient. The results are summarized in the table.Co-infections detected by the Karius Test in patients hospitalized with COVID-19 ConclusionOpen-ended, plasma-based NGS for mcfDNA provides a non-invasive method to assess for co-infections in critically ill patients with COVID-19. This report highlights the potential to increase diagnostic yield as well as to decrease the need for invasive procedures – and their attendant risks to patients and HCP – to obtain etiologic diagnoses to better inform antimicrobial therapy for superinfection. It also serves to highlight the variety of pathogens affecting these patients during the COVID-19 pandemic.DisclosuresWilliam V. La Via, MD, Karius (Employee) Sudeb Dalai, MD, Karius (Employee) Christiaan R. de Vries, MD, PhD, Karius (Consultant, Independent Contractor)Stanford University (Employee) Ann Macintyre, DO, Karius (Employee) Asim A. Ahmed, MD, Karius (Employee)

670. Rapid, Non-invasive Detection of Invasive Mycoplasma hominis Infection using the Karius Test, A Next-Generation Sequencing Test for Microbial Cell-free DNA in Plasma

Background Mycoplasma hominis is typically associated with genital infections in women and is a rare cause of musculoskeletal infections often in immunocompromised hosts. Diagnosis of invasive Mycoplasma hominis infections are difficult due to challenges in culturing these organisms. Molecular diagnostics require an index of suspicion which may not be present at the time of tissue sampling. Accurate, rapid diagnosis of Mycoplasma hominis infections are important for antibiotic management.MethodsTwo cases of invasive Mycoplasma hominis infections are presented in which the Karius test (KT) was used to make the diagnosis. The KT is a CLIA certified/CAP-accredited next-generation sequencing (NGS) plasma test that detects microbial cell-free DNA (mcfDNA). After mcfDNA is extracted and NGS performed, human reads are removed and remaining sequences are aligned to a curated database of > 1400 organisms. Organisms present above a statistical threshold are reported. Case review was performed for clinical correlation.ResultsA young woman with lupus nephritis status post renal transplant developed persistent fever with progressive multifocal culture-negative osteoarticular infection despite empiric ceftriaxone. An adolescent female presented with an ascending pelvic infection progressing to purulent polymicrobial peritonitis (see table) requiring surgical debridement and cefipime, metronidazole and micafungin therapy; her course was complicated by progressive peritonitis/abscesses. Karius testing detected high-levels of Mycoplasma hominis mcfDNA in both cases – at 3251 molecules/microliter (MPM) in the first case and 3914 MPM in the second case. The normal range of Mycoplasma hominis mcfDNA in a cohort of 684 normal adults is 0 MPM. The patients rapidly improved with atypical coverage with doxycycline and levofloxaxin.Clinical findings in 2 patients with M. hominis infection detected by the Karius Test ConclusionOpen-ended, plasma-based NGS for mcfDNA provides a rapid, non-invasive method to diagnose invasive Mycoplasma hominis infection. This case series highlights the potential to diagnose infections caused by fastidious pathogens to better inform antimicrobial therapy and achieve favorable outcomes.DisclosuresWilliam V. La Via, MD, Karius (Employee)

395. Rapid, Non-invasive Detection of Infection Using Plasma-based Next-Generation Sequencing for Microbial Cell-free DNA in Individuals Testing Negative for SARS-CoV-2 in a Pandemic Setting

BackgroundThe clinical presentation of patients with severe COVID-19 infection can be protracted and deteriorate to ARDS and multi-organ dysfunction with prolonged fever. As such, there is clinical overlap with many infectious diseases especially those that cause pneumonia. Due to of the prevalence of COVID-19 illness amidst the pandemic, concerns about testing sensitivity and the attendant risk to health care personnel (HCP) delivering care, patients are frequently tested multiple times to ascertain that they are SARS-CoV-2 free. Often, alternative diagnoses are not considered because some diagnostic modalities—such as bronchoalveolar lavage (BAL)—pose an unacceptable risk to the patient and/or HCP.MethodsWe interrogated plasma for microbial cell-free DNA from 58 patients who were known to be SARS-CoV-2 negative. Clinical information is taken from information submitted with the test requisition or obtained at the time of result reporting from clinical consultations with the ordering provider. In each case, a plasma sample was analyzed with the Karius Test (KT) which is a CLIA certified/CAP-accredited next-generation sequencing (NGS) plasma test designed to detect and quantify circulating microbial cell-free DNA (mcfDNA), which can assist with the diagnosis of deep-seated infections. After mcfDNA is extracted and NGS performed, human reads are removed and remaining sequences are aligned to a curated database of >1400 organisms. Organisms present above a statistical threshold are reported. The time to result is on average 24 hours from sample receipt.ResultsIn a subset of 20 samples, we found a broad range of pathogens. Pneumocystis jirovecii was the most common. These detections were unexpected in the majority of these patients. (see Table)Broad range of Karius detected pathogens (including fastidious bacteria, mycobacteria, fungi and viruses) ConclusionOpen-ended, plasma-based NGS for mcfDNA with the KT provides a rapid, non-invasive method to diagnose deep-seated infection like pneumonia. This broad-based test detected a wide range of pathogens – many unsuspected – in patients with severe pneumonia and other invasive infections during the COVID-19 pandemic. These detections highlight the utility of the tool; which allowed better management including de-escalation of SARS-CoV-2 testing and selection of appropriate antibiotic therapy for the unexpected diagnoses.DisclosuresWilliam V. La Via, MD, Karius (Employee) Sudeb Dalai, MD, Karius (Employee) Christiaan R. de Vries, MD, PhD, Karius (Consultant, Independent Contractor)Stanford University (Employee) Ann Macintyre, DO, Karius (Employee) Asim A. Ahmed, MD, Karius (Employee)

Development of Methods to Extract RNA From Archived Pediatric Needle Liver Biopsies to Produce Sequencing Data.

Genetic susceptibility has been proposed as etiopathogenic in several pediatric liver diseases including autoimmune hepatitis (AIH). High throughput sequencing (HTPS) has been applied to archived needle liver biopsies obtained from adults but rarely to pediatric biopsies. For conclusive diagnosis of AIH, most subjects have an initial formalin-fixed paraffin embedded (FFPE) needle liver biopsy that is eventually archived and may be stored for decades. Our goal was to develop methods to utilize tissue from archived needle liver biopsies for extraction of RNA sufficient to produce HTPS data. We extracted total RNA from 45 FFPE needle liver biopsy samples (24 AIH type 1 patients and 21 controls [ages 15_11 and 19_10]; biopsy storage time 0.5-20 years) and constructed cDNA libraries that were then sequenced on an Illumina HiSeq2000 platform. Forty (89%) of the libraries produced high-quality sequences for further analyses. The average number of sequences obtained per library from HTPS was 55,136,519 (range 14,914,291-184,027,499). There was a significant inverse relationship between the number of human reads obtained and the age of the specimen (P < 2_10_7). It was possible to classify more than 90% of the reads as known genes in samples that had been stored for less than 10 years. Archived needle liver biopsies can be used for sequence based interrogation of the etiologic origins of complex liver diseases of young subjects, such as AIH.

Development of a support vector machine learning and smart phone Internet of Things-based architecture for real-time sleep apnea diagnosis

BackgroundThe breathing disorder obstructive sleep apnea syndrome (OSAS) only occurs while asleep. While polysomnography (PSG) represents the premiere standard for diagnosing OSAS, it is quite costly, complicated to use, and carries a significant delay between testing and diagnosis.MethodsThis work describes a novel architecture and algorithm designed to efficiently diagnose OSAS via the use of smart phones. In our algorithm, features are extracted from the data, specifically blood oxygen saturation as represented by SpO2. These features are used by a support vector machine (SVM) based strategy to create a classification model. The resultant SVM classification model can then be employed to diagnose OSAS. To allow remote diagnosis, we have combined a simple monitoring system with our algorithm. The system allows physiological data to be obtained from a smart phone, the data to be uploaded to the cloud for processing, and finally population of a diagnostic report sent back to the smart phone in real-time.ResultsOur initial evaluation of this algorithm utilizing actual patient data finds its sensitivity, accuracy, and specificity to be 87.6%, 90.2%, and 94.1%, respectively.DiscussionOur architecture can monitor human physiological readings in real time and give early warning of abnormal physiological parameters. Moreover, after our evaluation, we find 5G technology offers higher bandwidth with lower delays ensuring more effective monitoring. In addition, we evaluate our algorithm utilizing real-world data; the proposed approach has high accuracy, sensitivity, and specific, demonstrating that our approach is very promising.ConclusionsExperimental results on the apnea data in University College Dublin (UCD) Database have proven the efficiency and effectiveness of our methodology. This work is a pilot project and still under development. There is no clinical validation and no support. In addition, the Internet of Things (IoT) architecture enables real-time monitoring of human physiological parameters, combined with diagnostic algorithms to provide early warning of abnormal data.

MegaPath: sensitive and rapid pathogen detection using metagenomic NGS data

BackgroundNext-generation sequencing (NGS) enables unbiased detection of pathogens by mapping the sequencing reads of a patient sample to the known reference sequence of bacteria and viruses. However, for a new pathogen without a reference sequence of a close relative, or with a high load of mutations compared to its predecessors, read mapping fails due to a low similarity between the pathogen and reference sequence, which in turn leads to insensitive and inaccurate pathogen detection outcomes.ResultsWe developed MegaPath, which runs fast and provides high sensitivity in detecting new pathogens. In MegaPath, we have implemented and tested a combination of polishing techniques to remove non-informative human reads and spurious alignments. MegaPath applies a global optimization to the read alignments and reassigns the reads incorrectly aligned to multiple species to a unique species. The reassignment not only significantly increased the number of reads aligned to distant pathogens, but also significantly reduced incorrect alignments. MegaPath implements an enhanced maximum-exact-match prefix seeding strategy and a SIMD-accelerated Smith-Waterman algorithm to run fast.ConclusionsIn our benchmarks, MegaPath demonstrated superior sensitivity by detecting eight times more reads from a low-similarity pathogen than other tools. Meanwhile, MegaPath ran much faster than the other state-of-the-art alignment-based pathogen detection tools (and compariable with the less sensitivity profile-based pathogen detection tools). The running time of MegaPath is about 20 min on a typical 1 Gb dataset.