Linking Genotype to Phenotype: Further Exploration of Mutations in SARS-CoV-2 Associated with Mild or Severe Outcomes

  1. Roshna Agarwal
  2. Tyler Leblond
  3. Erin M McAuley
  4. Ezekiel J Maier
  5. Martin Skarzynski
  6. Jameson D Voss
  7. Shanmuga Sozhamannan

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

We previously interrogated the relationship between SARS-CoV-2 genetic mutations and associated patient outcomes using publicly available data downloaded from GISAID in October 2020 [1]. Using high-level patient data included in some GISAID submissions, we were able to aggregate patient status values and differentiate between severe and mild COVID-19 outcomes. In our previous publication, we utilized a logistic regression model with an L1 penalty (Lasso regularization) and found several statistically significant associations between genetic mutations and COVID-19 severity. In this work, we explore the applicability of our October 2020 findings to a more current phase of the COVID-19 pandemic.

Here we first test our previous models on newer GISAID data downloaded in October 2021 to evaluate the classification ability of each model on expanded datasets. The October 2021 dataset (n=53,787 samples) is approximately 15 times larger than our October 2020 dataset (n=3,637 samples). We show limitations in using a supervised learning approach and a need for expansion of the feature sets based on progression of the COVID-19 pandemic, such as vaccination status. We then re-train on the newer GISAID data and compare the performance of our two logistic regression models. Based on accuracy and Area Under the Curve (AUC) metrics, we find that the AUC of the re-trained October 2021 model is modestly decreased as compared to the October 2020 model. These results are consistent with the increased emergence of multiple mutations, each with a potentially smaller impact on COVID-19 patient outcomes. Bioinformatics scripts used in this study are available at https://github.com/JPEO-CBRND/opendata-variant-analysis . As described in Voss et al. 2021, machine learning scripts are available at https://github.com/Digital-Biobank/covid_variant_severity .

Article activity feed

  1. Version published to 10.1101/2022.04.15.22273922v2 on medRxiv
  2. ScreenIT

    SciScore for 10.1101/2022.04.15.22273922: (What is this?)

    Please note, not all rigor criteria are appropriate for all manuscripts.

    Table 1: Rigor

    NIH rigor criteria are not applicable to paper type.

    Table 2: Resources

    Software and Algorithms
    SentencesResources
    Briefly, an export of raw GISAID SARS-CoV-2 data was curated using Nexstrain’s ncov-ingest shell scripts [2] and FASTA sequences were parsed from the data export using Python (version 3.8.10).
    Python
    suggested: (IPython, RRID:SCR_001658)
    Resulting VCF (Variant Call Format) files were merged using bcftools and annotated using SnpEff and filtered using SnpSift.
    SnpEff
    suggested: (SnpEff, RRID:SCR_005191)
    SnpSift
    suggested: (SnpSift, RRID:SCR_015624)
    ROC curves were plotted using Scikit-learn [4], and Matplotlib [5].
    Matplotlib
    suggested: (MatPlotLib, RRID:SCR_008624)
    AUC confidence intervals, P-values, and diagnostic odds ratios (OR) were calculated using NumPy [9] for each of the five logistic regression models.
    NumPy
    suggested: (NumPy, RRID:SCR_008633)
    The Scikit-learn implementation of logistic regression does not provide ORs or P-values for individual variables.
    Scikit-learn
    suggested: (scikit-learn, RRID:SCR_002577)
    ORs and Chi-square test P-values for the association of mutations with Severe and Mild outcomes (Figure 5) were calculated from mutation count data using Statsmodels and SciPy respectively [8].
    SciPy
    suggested: (SciPy, RRID:SCR_008058)

    Results from OddPub: Thank you for sharing your code and data.

    Results from LimitationRecognizer: We detected the following sentences addressing limitations in the study:
    The utilization of supervised learning machine learning poses a limitation in our analysis. Since labeled outcomes are required to train these models, the number of samples available for training is reduced by 99% (53,787 of 4,646,285). In addition, machine learning models trained on older samples may not be sufficiently exposed to new mutations. For example, while many of the more than 50 mutations present in Omicron were observed previously in other variants of concern, some Omicron mutations were rare or previously unobserved and many previously observed mutations hadn’t co-occurred in the same samples [16]. Supervised machine learning models cannot effectively utilize previously unobserved mutations and mutations combinations because parameters have not been fit for these features. A promising approach for addressing these limitations is semi-supervised learning. This machine learning approach uses both labeled data and unlabeled data for model training. Semi-supervised learning may outperform supervised learning approaches when the amount of unlabeled data is much larger than labeled data [17]. Within the field of genomics, recent example uses of semi-supervised learning include microRNA classification [18], somatic genomic variant classification [19], and identify disease associated genes [20].

    Results from TrialIdentifier: No clinical trial numbers were referenced.

    Results from Barzooka: We did not find any issues relating to the usage of bar graphs.

    Results from JetFighter: We did not find any issues relating to colormaps.

    Results from rtransparent:
    • Thank you for including a conflict of interest statement. Authors are encouraged to include this statement when submitting to a journal.
    • Thank you for including a funding statement. Authors are encouraged to include this statement when submitting to a journal.
    • No protocol registration statement was detected.

    Results from scite Reference Check: We found no unreliable references.

    About SciScore

    SciScore is an automated tool that is designed to assist expert reviewers by finding and presenting formulaic information scattered throughout a paper in a standard, easy to digest format. SciScore checks for the presence and correctness of RRIDs (research resource identifiers), and for rigor criteria such as sex and investigator blinding. For details on the theoretical underpinning of rigor criteria and the tools shown here, including references cited, please follow this link.

  3. Version published to 10.1101/2022.04.15.22273922v1 on medRxiv