The COV-IRT microbial subgroup is actively analyzing publicly available bronchoalveolar lavage fluid (BALF) metatranscriptomes to search for correlations between microbial genes, functions, and taxa in COVID-19 patients (e.g. utilizing data from PMIDs: 32020836, 32015507, 32129843, 32228226, 31836714, 30417108, 30389465, https://www.biorxiv.org/content/10.1101/2020.01.24.919183v2, https://osf.io/7nrd3/wiki/home/). Understanding the potential role the human microbiome plays in the susceptibility to and severity of COVID-19 is an important consideration to improve the precision of COVID-19 diagnosis, outcome prediction, and treatment options. Respiratory tract microbiomes in COVID-19 patients have expressed a reduced diversity of alphaproteobacteria and an increased abundance of respiratory pathogens and oral and upper respiratory commensal bacteria. This is similar to the microbiome of virus-like community-acquired pneumonia patients (PMID: 32463434, 32129843). We hypothesize that specific bacterial species, such as those in the Prevotella genus, are potential pathogens that thrive in COVID-19-induced inflammatory environments at infected mucosal sites (PMID: 28542929). To assess this hypothesis and leverage broader COV-IRT expertise in human disease inflammatory pathways, we are collaborating with the COV-IRT Inflammation subgroup to review supporting data and the presence of Prevotella human-response signatures in COVID-19 patients (e.g., up-regulated human transcripts in Th17-mediated mucosal inflammation). In addition to detecting taxonomic shifts that may be driven by COVID-19 microbiome dysbiosis, we are comparing the functional profile of microbial communities through the analysis of predicted proteins and gene ontology terms assigned to sequence reads. A previous study with a small number of COVID-19 BALF samples (N=8) suggested that the disease causes airway microbiomes to have increased vitamin (e.g., B6, folate, lipoic acid), drug, nucleotide (e.g., purine, pyrimidine), and energy (e.g., methane, nitrogen, sulfur) metabolism, yet decreased amino acid and carbohydrate metabolism (https://www.biorxiv.org/content/10.1101/2020.05.01.073171v2). We are building on the initial findings of this study by analyzing microbial gene ontology in these and additional public BALF metatranscriptomes. In a future collaboration with the COV-IRT Mitochondria and Neurology subgroups, we will evaluate the hypothesis that COVID-19 gut-brain-microbiome functional changes are consistent with a response to host reactive oxygen species production and specific host mitochondrial DNA variants (PMID: 31266851).

The Inflammation subgroup was formed to investigate links between altered immune signaling in COVID-19 patients with known processes of inflammation. We are exploring novel hypotheses derived for the role of the inflammasome, which is linked to both lung epithelial and immune cells in the pathogenesis of pulmonary disease in COVID-19 patients. While it is known that a cytokine storm evolves in many of these patients, the role of the inflammasome in initiating a cytokine storm in coordination with viral entry is not understood.When overactive during infection, this pathway can create severe autoimmune-like inflammation, triggering epithelial cell death even for neighboring cells not harboring the pathogen.

Recent studies quantifying defined interactions through genomics data analyses suggest novel links between induction of key signaling components of the inflammasome and induced defects in DNA repair.  Such defects include Fanconi anemia defects and those that affect homologous recombination (PMID31591209). We are currently evaluating the hypothesis that this phenomenon may help drive the severity of lung pathogenesis in COVID-19 patients. We also aim to understand how age and sex impact the inflammasome to drive the severity of the disease. The ultimate aim of this group is to use a derivation of the above biology to devise a clinically useful serum/plasma protein signature that may be used to monitor disease status, project the course of COVID-19 patients with severe lung disease, predict early infection in COVID-19 patients most likely to progress to severe disease and identify novel therapeutics to treat these patients. Furthermore recent evidence suggests anti-inflammatory strategies may be effective as a potential treatment method for COVID-19. (PMID: 32352871) The therapeutic benefit of pharmacological modulation of a key inflammatory lipid pathway by dexamethasone (PMID: 32610262) provides further rationale for investigating the lipid signaling in COVID-19 patients.

The overall function of the miRNA/non-coding RNA subgroup is to focus on novel biology related to both SARS-COV-2 and the host. Currently, there is limited knowledge on the impact of non-coding RNAs on SARS-COV-2 and the host after infection. Previous studies with other viruses, including coronaviruses, have shown that viruses can co-opt host miRNAs to enhance immune system evasion and viral replication (PMID: 28555130, PMID: 27373545). In addition, it is becoming apparent that the resulting infection due to COVID-19 in patients causes a systemic impact on the entire body as is evident by the growing list of symptoms during COVID-19 infection. A single miRNA can regulate hundreds of mRNAs and circulate throughout the body to contribute to the systemic impact of specific diseases (PMID: 30123182). We hypothesize that one or more miRNAs contribute to the downstream effects in the patient caused by SARS-COV-2 infection. The goals of the group are to identify the specific miRNAs associated with SARS-COV-2, determine if these miRNAs can be utilized as a novel biomarker/detection method for COVID-19, design a potential miRNA-based therapeutic/vaccine, and implement a miRNA-based COVID-19 therapy. This subgroup is currently pursuing its goals by utilizing predictive miRNA analysis on RNA-seq data and analysis on miRNA-seq data on COVID-19 from publicly available databases. The data that has been generated to date has compared predicted miRNAs from existing RNA-seq and related to specific regions of SARS-CoV-2. In addition, predicted miRNAs are being compared to viral load in patients to determine potential ideal miRNA candidate to target for therapeutics. We have also started to test potential miRNAs based targets in both in vitro and in vivo COVID-19 models. Lastly, this group is exploring the regulatory roles of other non-coding RNAs in COVID-19 infection, such as long noncoding RNAs (lncRNAs) and circular RNAs (circRNAs), with similar techniques.

The radiology and imaging subgroup is primarily interested in answering the research question: Can we predict clinical outcomes and patient risk of COVID-19 emergency room patients using ordinary and inexpensive clinical data and X-Ray examination? To answer this question we are collecting clinical data and chest X-rays from patients admitted to the ER in Milan hospitals for any lung illness. We will start by separately processing clinical and imaging data. Clinical data will be analyzed through statistical (PMID: 30034025) and machine learning (ML) methods such as Random Forest, logistic regression, and semi-supervised graph-based algorithms (PMID: 29045534, PMID: 28376093, PMID: 32107391,); X-ray images through deep neural networks (PMID: 30134902, PMID: 31000728), trained by transfer learning (PMID: 29474911) in case of image datasets of limited cardinality, to identify their separate potential in determining COVID-19 risk assessment. The models developed with the two types of data (clinical and X-ray) will then be merged into a multi-modal classifier for patients’ risk stratification. We plan to extend the bi-modal imaging/clinical prediction system to include other sources of data (e.g. omics and EHR data) using integrated ML approaches, in collaboration with other COV-IRT subgroups, to further improve the performance of the ML-based prediction system. To gain trust and increase applicability, all the classification results will be supported by explanations (PMID: 30023379, PMID: 31322793), which will report the reasons behind the predicted patient risk and will highlight the clinical variables or lung areas identified as being the main factors causing the specific prediction. To this aim, we will develop and apply XAI (Explainable Artificial Intelligence) methods to provide human-understandable motivations underlying machine learning predictions, thus “opening the black-box” prediction models (L. H. Gilpin et al IEEE 5th International Conference on Data Science and Advanced Analytics, 2018, doi: 10.1109/DSAA.2018.00018)

This subgroup takes an interdisciplinary analysis and modeling approach to understand host-viral dynamics and interactions with environmental factors. Our goal is to bring together data across all the COV-IRT subgroups to use statistical analyses, mechanistic models, agent-based models, and machine learning methods to address the research question posed by the subgroup. Currently, this subgroup is: 1) developing methods to evaluate anomalies in reporting of COVID-19 deaths, and how this could impact predictive models; 2) evaluating the influence of human behavior (e.g., activity, interpersonal contacts) on virus spread; 3) evaluating the impact of environmental factors on the transmission of the disease, such as population density and weather conditions; 4) establishing corrected and accurate transmission and mortality rates to relate to virus strains geographically 5) using deep learning to detect COVID-19 from medical imaging data; and 6) conducting a large-scale analysis of COVID-19 resources on Github to identify patterns in collaboration and types of resources available to assess the form and state of “open science”. We will collaborate with other subgroups to integrate EHR, medical imaging, genomics, transcriptomics, and epidemiological data to explore host-viral interactions, predict future outbreaks, detect biomarkers, and design therapeutics.

Our goal is to develop new avenues for precision diagnostics and therapeutics to mitigate COVID-19. The COVID-19 pandemic affects populations differently. Notably, in the US and Britain, ethnic minorities including people of African descent have disproportionately high morbidity and mortality (PMID: 32543702, PMID: 32557713). Many factors are likely responsible (PMID: 32664879, PMID: 32543702, PMID: 32557713, https://www.biorxiv.org/content/10.1101/2020.06.09.143271v3). Our team is focused specifically on molecular factors that may be related to this disparity in COVID-19 severity.

Gene expression is the interaction of environment by genetics, and as such can reveal clues about the molecular milieu that is caused by either. One approach we are taking is to computationally mine raw data from aggregated ‘omics studies on COVID-19 disease as these become publicly available. As of July 31, 2020, data from 11 single cell and bulk RNA studies are available (https://github.com/urmi-21/COVID-19-RNA-Seq-datasets). We will integrate these data with the metadata (information about the technical samples and the individuals sampled, such as race/ethnicity, age, gender, disease severity, time since infection, and obesity) to identify prognostic markers of COVID-19 and to develop a new understanding of how genetics may affect disease progression and how these may vary across populations.

Our analysis of RNA-Seq data is identifying transcripts of known protein-coding genes as well as transcripts representing novel species-specific protein-coding genes (often called “orphan” or “de novo” genes”. Some orphan genes, if of very recent origin, may be population-specific. Orphan genes are thought to be a key enabler of speciation (PMID: 23949544, PMID: 25151064).  Of those functionally characterized to date, many are implicated in pathogenesis (PMID: 29878511, PMID: 10706094) or immune function (PMID: 30075701, PMID: 30635418). The known genes and novel orphan genes that we identify as being associated with race will be functionally assessed in silico and selected genes will be subject to wet-lab experimentation. Key in silico tools that will facilitate our analysis include MetaOmGraph, a Java software that empowers flexible, interactive statistical analysis and exploration of aggregated data sets by domain experts (PMID: 31956905), pyrpipe, a Python package for reproducibility and flexibility in RNA-Seq analysis, fagin, an R tool (PMID: 31455236), phylostratr, for automated inference of gene age (PMID: 30873536), and other packages.

The pipeline subgroup develops consensus pipelines for processing raw RNA sequence data to generate alignment, quantification, and variant call data, which will subsequently be used by other COV-IRT subgroups for downstream analyses. This ensures that all COV-IRT subgroups performing downstream analyses start with data that was processed the same way, thereby eliminating biases that may arise from using different tools to generate processed data prior to downstream analyses.