Genomic Surveillance of COVID-19 Variants With Language Models and Machine Learning
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (ScreenIT)
Abstract
The global efforts to control COVID-19 are threatened by the rapid emergence of novel SARS-CoV-2 variants that may display undesirable characteristics such as immune escape, increased transmissibility or pathogenicity. Early prediction for emergence of new strains with these features is critical for pandemic preparedness. We present Strainflow , a supervised and causally predictive model using unsupervised latent space features of SARS-CoV-2 genome sequences. Strainflow was trained and validated on 0.9 million sequences for the period December, 2019 to June, 2021 and the frozen model was prospectively validated from July, 2021 to December, 2021. Strainflow captured the rise in cases 2 months ahead of the Delta and Omicron surges in most countries including the prediction of a surge in India as early as beginning of November, 2021. Entropy analysis of Strainflow unsupervised embeddings clearly reveals the explore-exploit cycles in genomic feature-space, thus adding interpretability to the deep learning based model. We also conducted codon-level analysis of our model for interpretability and biological validity of our unsupervised features. Strainflow application is openly available as an interactive web-application for prospective genomic surveillance of COVID-19 across the globe.
Article activity feed
-
-
-
-
-
SciScore for 10.1101/2021.05.25.445601: (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 Sentences Resources We used Boruta [17], a wrapper algorithm around randomforest to select the important blip dimension for the prediction of subsequent 1 month’s new cases Borutasuggested: (Boruta, RRID:SCR_016234)Results from OddPub: We did not detect open data. We also did not detect open code. Researchers are encouraged to share open data when possible (see Nature blog).
Results from LimitationRecognizer: We detected the following sentences addressing limitations in the study:There are few limitations to our study. Although the embeddings indicate 3-mers with high weights, these do not indicate the …
SciScore for 10.1101/2021.05.25.445601: (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 Sentences Resources We used Boruta [17], a wrapper algorithm around randomforest to select the important blip dimension for the prediction of subsequent 1 month’s new cases Borutasuggested: (Boruta, RRID:SCR_016234)Results from OddPub: We did not detect open data. We also did not detect open code. Researchers are encouraged to share open data when possible (see Nature blog).
Results from LimitationRecognizer: We detected the following sentences addressing limitations in the study:There are few limitations to our study. Although the embeddings indicate 3-mers with high weights, these do not indicate the position where the 3-mer change may have happened in the genome. This is because low-dimensional embeddings do not preserve the positional encoding of words. However, we plan to evaluate advanced approaches such as complex valued word embeddings with positional encodings [36] and transformer models such as BERT [37][38]. The latter are considered expensive and data-hungry models and it will remain to be evaluated if the gain of positional information may be countered by the loss of prediction accuracy for forecasting new cases in the future. However, we believe that the availability of sequences for a wide variety of viral pathogens presents an exciting opportunity to train data-hungry models that may be able to transfer insights across pathogens and yet remain interpretable. Another limitation of our study is the relatively small number of samples that were used to construct the supervised predictive models. This is to eliminate the sampling bias that may arise while building the supervised models as some countries had a disproportionately higher number of samples submitted to GISAID. However, the unsupervised embeddings and temporal cross-correlations were learned upon the full datasets and these presented clear patterns in DoCs and significant cross correlations with caseloads. Nevertheless, it is important for our models to receive at least 30 sampl...
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.
-