COVID-19 and Artificial Intelligence: An Approach to Forecast the Severity of Diagnosis

(1) Background: The new SARS-COV-2 pandemic overwhelmed intensive care units, clinicians, and radiologists, so the development of methods to forecast the diagnosis’ severity became a necessity and a helpful tool. (2) Methods: In this paper, we proposed an artificial intelligence-based multimodal approach to forecast the future diagnosis’ severity of patients with laboratory-confirmed cases of SARS-CoV-2 infection. At hospital admission, we collected 46 clinical and biological variables with chest X-ray scans from 475 COVID-19 positively tested patients. An ensemble of machine learning algorithms (AI-Score) was developed to predict the future severity score as mild, moderate, and severe for COVID-19-infected patients. Additionally, a deep learning module (CXR-Score) was developed to automatically classify the chest X-ray images and integrate them into AI-Score. (3) Results: The AI-Score predicted the COVID-19 diagnosis’ severity on the testing/control dataset (95 patients) with an average accuracy of 98.59%, average specificity of 98.97%, and average sensitivity of 97.93%. The CXR-Score module graded the severity of chest X-ray images with an average accuracy of 99.08% on the testing/control dataset (95 chest X-ray images). (4) Conclusions: Our study demonstrated that the deep learning methods based on the integration of clinical and biological data with chest X-ray images accurately predicted the COVID-19 severity score of positive-tested patients.

EP.WE.372Has COVID-19 affected the attention surgical papers are receiving?

Aim Unsurprisingly, much of the medical profession has focussed on treating covid-19 over the past year often to the detriment of other pathologies. Engagement with academic literature may follow a similar trend; this paper used standard social media attention and citations metrices to assess whether the pandemic has affected engagement with surgical literature. Method Twitter Mentions and Mendeley Readers Data were retrieved for all papers published in Annals of Surgery, BJS and JAMASurgery between January 2019 and October 2020. The non-parametric Mann-Whitney U test was used to compare Twitter Mentions and Mendeley Readers for COVID surgical vs non-COVID surgical publications (in 2020) and all surgical papers published before and after the advent of COVID-19. A control database of all papers published in NEJM, BMJ and Lancet over the same period of time was also created. Results There was no difference in Twitter mentions between COVID-19 and non-COVID-19 papers (p-value 0.604); however there were significantly more Mendeley readers of COVID-19 papers than non-COVID-19 papers in 2020 (55 vs 5 median readers, p < 0.001). Surgical papers published in 2020 received significantly fewer Twitter mentions than those published in 2019 (15.5 vs 27, p < 0.001). Analysis of the non-surgical control dataset revealed the opposite trend; papers published in 2020 received significantly more attention than those published in 2019 (39 vs 30 median Twitter citations, p < 0.001). Conclusion Surgical papers published during the COVID-19 pandemic received significantly fewer social media mentions. Such reduced visibility has the potential to affect future citation metrics and dissemination of surgical knowledge

Performance of Two Different Flight Configurations for Drone-Borne Magnetic Data

The compensation of magnetic and electromagnetic interference generated by drones is one of the main problems related to drone-borne magnetometry. The simplest solution is to suspend the magnetometer at a certain distance from the drone. However, this choice may compromise the flight stability or introduce periodic data variations generated by the oscillations of the magnetometer. We studied this problem by conducting two drone-borne magnetic surveys using a prototype system based on a cesium-vapor magnetometer with a 1000 Hz sampling frequency. First, the magnetometer was fixed to the drone landing-sled (at 0.5 m from the rotors), and then it was suspended 3 m below the drone. These two configurations illustrate endmembers of the possible solutions, favoring the stability of the system during flight or the minimization of the mobile platform noise. Drone-generated noise was filtered according to a CWT analysis, and both the spectral characteristics and the modelled source parameters resulted analogously to that of a ground magnetic dataset in the same area, which were here taken as a control dataset. This study demonstrates that careful processing can return high quality drone-borne data using both flight configurations. The optimal flight solution can be chosen depending on the survey target and flight conditions.

Host genome analysis of structural variations by Optical Genome Mapping provides clinically valuable insights into genes implicated in critical immune, viral infection, and viral replication pathways in patients with severe COVID-19.

Background: The varied clinical manifestations and outcomes in patients with SARS-CoV-2 infections implicate a role of host-genetics in the predisposition to disease severity. This is supported by evidence that is now emerging, where initial reports identify common risk factors and rare genetic variants associated with high risk for severe/ life-threatening COVID-19. Impressive global efforts have focused on either identifying common genetic factors utilizing short-read sequencing data in Genome-Wide Association Studies (GWAS) or whole-exome and genome studies to interrogate the human genome at the level of detecting single nucleotide variants (SNVs) and short indels. However, these studies lack the sensitivity to accurately detect several classes of variants, especially large structural variants (SVs) including copy number variants (CNVs), which account for a substantial proportion of variation among individuals. Thus, we investigated the host genomes of individuals with severe/life-threatening COVID-19 at the level of large SVs (500bp-Mb level) to identify events that might provide insight into the inter-individual clinical variability in clinical course and outcomes of COVID-19 patients. Methods: Optical genome mapping using Bionano Saphyr system was performed on thirty-seven severely ill COVID-19 patients admitted to intensive care units (ICU). To extract candidate SVs, three distinct analyses were undertaken. First, an unbiased whole-genome analysis of SVs was performed to identify rare/unique genic SVs in these patients that did not appear in population datasets to determine candidate loci as decisive predisposing factors associated with severe COVID-19. Second, common SVs with a population frequency filter was interrogated for possible association with severe COVID-19 based on literature surveys. Third, genome-wide SV enrichment in severely ill patients versus the general population was investigated by calculating odds ratios to identify top-ranked genes/loci. Candidate SVs were confirmed using qPCR and an independent bioinformatics tool (FaNDOM). Results: Our patient-centric investigation identified 11 SVs involving 38 genes implicated in three key host-viral interaction pathways: (1) innate immunity and inflammatory response, (2) airway resistance to pathogens, and (3) viral replication, spread, and RNA editing. These included seven rare/unique SVs (not present in the control dataset), identified in 24.3% (9/37) of patients, impacting up to 31 genes, of which STK26 and DPP4 are the most promising candidates. A duplication partially overlapping STK26 was corroborated with data showing upregulation of this gene in severely ill patients. Further, using a population frequency filter of less than 20% in the Bionano control dataset, four SVs involving seven genes were identified in 56.7% (21/37) of patients. Conclusion: This study is the first to systematically assess and highlight SVs' potential role in the pathogenesis of COVID-19 severity. The genes implicated here identify novel SVs, especially STK26, and extend previous reports involving innate immunity and type I interferon response in the pathogenesis of COVID-19. Our study also shows that optical genome mapping can be a powerful tool to identify large SVs impacting disease outcomes with split survival and add valuable genomic information to the existing sequencing-based technology databases to understand the inter-individual variability associated with SARS-CoV-2 infections and COVID-19 mortality.

Visual Identification of Inconsistency in Pattern

The visual identification of inconsistencies in patterns is an area in computing that has been understudied. While pattern visualisation exposes the relationships among identified regularities, it is still very important to identify inconsistencies (irregularities) in identified patterns. The significance of identifying inconsistencies for example in the growth pattern of children of a particular age will enhance early intervention such as dietary modifications for stunted children. It is described in this chapter, the need to have a system that identifies inconsistencies in identified pattern of a dataset. Also, techniques that enable the visual identification of inconsistencies in patterns such as fault tolerance and colour coding are described. Two approaches are presented in this chapter for visualising inconsistencies in patterns namely; visualising inconsistencies in objects with many attribute values and visual comparison of an investigated dataset with a case control dataset. These approaches are associated with tools which were developed by the authors of this chapter: Firstly, ConTra which allows its users to mine and analyse the contradictions in attribute values whose data does not abide by the mutual exclusion rule of the dataset. Secondly, Datax which mines missing data; enables the visualisation of the missingness and the identification of the associated patterns. Finally, WellGrowth which explores Children’s growth dataset by comparing an investigated dataset (data obtained from a Primary Health Centre) with a case control dataset (data from the website of World Health Organisation). Instances of inconsistencies as discovered in the explored datasets are discussed.

Investigating hospital Mycobacterium chelonae infection using whole genome sequencing and hybrid assembly

Mycobacterium chelonae is a rapidly growing nontuberculous mycobacterium that is a common cause of nosocomial infections. Here we describe investigation of a possible nosocomial transmission of M. chelonae at the Hospital of the University of Pennsylvania (HUP). M. chelonae strains with similar high-level antibiotic resistance patterns were isolated from two patients who developed post-operative infections at HUP in 2017, suggesting a possible point source infection. The isolates, along with other clinical isolates from other patients, were sequenced using the Illumina and Oxford Nanopore technologies. The resulting short and long reads were hybrid assembled into draft genomes. The genomes were compared by quantifying single nucleotide variants in the core genome and assessed using a control dataset to quantify error rates in comparisons of identical genomes. We show that all M. chelonae isolates tested were highly dissimilar, as indicated by high pairwise SNV values, consistent with environmental acquisition and not a nosocomial point source. Our control dataset determined a threshold for evaluating identity between strains while controlling for sequencing error. Finally, antibiotic resistance genes were predicted for our isolates, and several single nucleotide variants were identified that have the potential to modulated drug resistance.

Going against the norm: validation of a novel alternative to brain SPECT normative datasets

Aim: Quantitative analysis of brain single photon emission computed tomography (SPECT) perfusion imaging is dependent on normative datasets that are challenging to produce. This study investigated the combination of SPECT neuroimaging from a large clinical population rather than small numbers of controls. The authors hypothesized this “population template” would demonstrate noninferiority to a control dataset, providing a viable alternative for quantifying perfusion abnormalities in SPECT neuroimaging. Methods: A total of 2, 068 clinical SPECT scans were averaged to form the “population template”. Validation was three-fold. First, the template was imported into SPECT brain analysis software, MIMneuro®, and compared against its control dataset of 90 individuals through its region and cluster analysis tools. Second, a cohort of 100 cognitively impaired subjects was evaluated against both the population template and MIMneuro®’s normative dataset to compute region-based metrics. Concordance and intraclass correlation coefficients, mean square deviations, total deviation indices, and limits of agreement were derived from these data to measure agreement and test for noninferiority. Finally, the same patients were clinically read in CereMetrix® to confirm that expected perfusion patterns appeared after comparison to the template. Results: MIMneuro®’s default threshold for normality is ± 1.65 z-score and this served as our noninferiority margin. Direct comparison of the template to controls produced no regions that exceeded this threshold and all clusters identified were far from statistically significant. Agreement measures revealed consistency between the softwares and that CereMetrix® results were noninferior to MIMneuro®, albeit with proportional bias. Visual analysis also confirmed that expected perfusion patterns appeared when individual scans were compared to the population template within CereMetrix®. Conclusions: The authors demonstrated a population template was noninferior to a smaller control dataset despite inclusion of abnormal scans. This suggests that our patient-based population template can serve as an alternative for identifying and quantifying perfusion abnormalities in brain SPECT.

Troll and Divide: The Language of Online Polarization

Political polarization, or the ideological distance between the political left and right, has grown steadily in recent decades. There is a rising concern over the role that ‘bad actors’ or trolls may play in polarization in online networks. In this research, we examine the processes by which trolls may sow intergroup conflict through polarizing rhetoric. We developed a dictionary to gauge online polarization by measuring language associated with communications that display partisan bias in their diffusion. We validated the polarized language dictionary in three different contexts and across multiple time periods. We found the polarization dictionary made out-of-set predictions, generalized to new political contexts (#BlackLivesMatter), and predicted partisan differences in public polls about COVID-19. Then we analyzed 383,510 tweets from a known Russian troll source (the Internet Research Agency) and found that their use of polarized language has increased over time. We also compared troll tweets from 3 different countries (N = 798,33) and found that they all utilize more polarized language on average than a control dataset of tweets from regular Americans (N = 1,507,300) and trolls have dramatically increased their use of polarized rhetoric over time. These results illustrate how trolls leverage polarized language. We also provide an open-source, simple tool for exploration of polarized communications on social media.

GDASC: a GPU parallel-based web server for detecting hidden batch factors

Summary We developed GDASC, a web version of our former DASC algorithm implemented with GPU. It provides a user-friendly web interface for detecting batch factors. Based on the good performance of DASC algorithm, it is able to give the most accurate results. For two steps of DASC, data-adaptive shrinkage and semi-non-negative matrix factorization, we designed parallelization strategies facing convex clustering solution and decomposition process. It runs more than 50 times faster than the original version on the representative RNA sequencing quality control dataset. With its accuracy and high speed, this server will be a useful tool for batch effects analysis. Availability and implementation http://bioinfo.nankai.edu.cn/gdasc.php. Contact [email protected] Supplementary information Supplementary data are available at Bioinformatics online.