The COVID 19 Enigma: AI Models as the Saviour
The virus continues to outsmart the world through a complex interplay of infection, mortality, immunity and reinfection. The disease imposed a threat on mankind by diverse clinical presentation, controversial evidence for treatment, fast-tracked vaccine development and unclear systemic implications. Artificial intelligence as the saviour... (https://www.cmu.edu/iii/about/news/2020/covid-ai-predictions.html).
The AI approach: Genomic Surveillance
It means tracking pathogens (bacteria, virus, any other disease-causing organism) using genomic sequencing. By looking at the sequence for a pathogen, the evolution can be tracked to note any change in it impacting its biological properties. Biological sequences contain a plethora of information that can be exploited for genomic surveillance.
The Discovery: Strainflow
An epidemiological early warning system to predict new caseloads in various countries The spike protein latent space representation learned by Strainflow model could be used as a proxy to capture the spatiotemporal diversity in the emerging SARS-CoV-2 strains across different countries. (https://www.frontiersin.org/articles/10.3389/fgene.2022.858252/full)
What is Strainflow?
"One of a kind" early warning system for emergence of new variants of concern and case surges. It is a supervised and causally predictive model using unsupervised latent space features of SARS-CoV-2 genome sequences. Towards this, Strainflow was trained and validated on 0.9 million sequences until June 2021 and counting more…
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.
An approach for analyzing the emerging strains based on the latent space of spike protein coding nucleotide sequences. Nucleotide sequences were chosen instead of proteins to capture and track the variations that may not have immediate functional consequences.
How does it work?
- Data-driven, de-novo approach.
- Uses complex mixtures for predictions
- Smart approach bypassing the need for expert understanding of the effects of individual mutations.
- Making easy the difficult task of providing simultaneous attention to many information pieces such as multiple codons.
- An accurate sense of the sharpness of an infection surge.
- Predicting the probability of a wave with a two-month lead time.
- Providing enough time for the healthcare systems to be prepared.
An interactive publicly available web application, created using the strainflow model and is updated on a monthly basis http://strainflow.tavlab.iiitd.edu.in/
Unmasking COVID 19 trends
Although the Strainflow approach does not predict the actual number, we do get an accurate sense of how sharp the surge might be. With successful predictions, prospectively validated over multiple surges, the Strainflow model has proven to be effective for predicting whether there is a likely surge with a two-month lead time, which could help the healthcare systems to be prepared.
Prof. Tavpritesh Sethi
Associate Professor, Computational Biology, IIITD
This AI/ML based strain prediction model uses change in genomic sequencing to provide early warning signals to policymakers/healthcare officials about a potential uptick in COVID case up to two months in advance, allowing sufficient time for pandemic preparedness. This model can be adapted to any infectious disease and is quite a potent tool for administrators and healthcare agencies to manage community health.