1. Guest, the 'News in Brief' for the week beginning 12th July 2021 is here.
    Dismiss Notice
  2. Welcome! To read the Core Purpose and Values of our forum, click here.
    Dismiss Notice
  3. Contribute to feedback on the CDC Evidence Review, for more details click here
    Dismiss Notice

Immune-Based Prediction of COVID-19 Severity and Chronicity Decoded Using Machine Learning Patterson et al. 2021

Discussion in 'BioMedical ME/CFS Research' started by Jaybee00, Jul 21, 2021 at 6:54 PM.

  1. Jaybee00

    Jaybee00 Senior Member (Voting Rights)

    Likes Received:

    Expression of CCR5 and its cognate ligands have been implicated in COVID-19 pathogenesis, consequently therapeutics directed against CCR5 are being investigated. Here, we explored the role of CCR5 and its ligands across the immunologic spectrum of COVID-19. We used a bioinformatics approach to predict and model the immunologic phases of COVID so that effective treatment strategies can be devised and monitored.

    We investigated 224 individuals including healthy controls and patients spanning the COVID-19 disease continuum. We assessed the plasma and isolated peripheral blood mononuclear cells (PBMCs) from 29 healthy controls, 26 Mild-Moderate COVID-19 individuals, 48 Severe COVID-19 individuals, and 121 individuals with post-acute sequelae of COVID-19 (PASC) symptoms. Immune subset profiling and a 14-plex cytokine panel were run on all patients from each group.

    B-cells were significantly elevated compared to healthy control individuals (P<0.001) as was the CD14+, CD16+, CCR5+ monocytic subset (P<0.001). CD4 and CD8 positive T-cells expressing PD-1 as well as T-regulatory cells were significantly lower than healthy controls (P<0.001 and P=0.01 respectively). CCL5/RANTES, IL-2, IL-4, CCL3, IL-6, IL-10, IFN-γ, and VEGF were all significantly elevated compared to healthy controls (all P<0.001). Conversely GM-CSF and CCL4 were in significantly lower levels than healthy controls (P=0.01). Data were further analyzed and the classes were balanced using SMOTE. With a balanced working dataset, we constructed 3 random forest classifiers: a multi-class predictor, a Severe disease group binary classifier and a PASC binary classifier. Models were also analyzed for feature importance to identify relevant cytokines to generate a disease score. Multi-class models generated a score specific for the PASC patients and defined as S1 = (IFN-γ + IL-2)/CCL4-MIP-1β. Second, a score for the Severe COVID-19 patients was defined as S2 = (IL-6+sCD40L/1000 + VEGF/10 + 10*IL-10)/(IL-2 + IL-8).

    Severe COVID-19 patients are characterized by excessive inflammation and dysregulated T cell activation, recruitment, and counteracting activities. While PASC patients are characterized by a profile able to induce the activation of effector T cells with pro-inflammatory properties and the capacity of generating an effective immune response to eliminate the virus but without the proper recruitment signals to attract activated T cells.
    Last edited by a moderator: Jul 21, 2021 at 11:13 PM
  2. Snow Leopard

    Snow Leopard Senior Member (Voting Rights)

    Likes Received:
    This is interesting, but requires further replication.
    In particular, the machine learning sorting processes need to be tested in independent studies.
    (such as the “PASC Score” defined as S1 = (IFN-γ + IL-2)/CCL4-MIP-1β), where S1 > 0.5 defines a PASC case)

    The stated specificity/sensitivity of 100% and 97% compared to healthy controls and Mild-Moderate acute COVID cases sounds impressive, but we know that results from machine learning algorithms are often over-fitted to the data. Some severe acute COVID cases were also positive according to that criteria, but that is no big deal.

    I'm also a little bothered by the (albeit common) assumption that slightly elevated cytokines = inflammation. When increased cytokines (or monocyte activation) is not the same as inflammation.
    Last edited: Jul 22, 2021 at 8:53 AM
    alktipping, Michelle and obeat like this.
  3. Hutan

    Hutan Moderator Staff Member

    Likes Received:
    New Zealand
    There was no group of people who have had Covid-19 but who don't have lingering symptoms. The healthy controls had to be seronegative.

    Being able to identify people who have never had Covid-19 from those who have recently had Covid-19 isn't as exciting as being able to find a difference between those who have Long Covid and those who don't after Covid-19.
    alktipping, Michelle, rvallee and 2 others like this.

Share This Page