Genomic Surveillance of COVID-19 Variants With Language Models and Machine Learning
Problem
Public health needs early prediction of new SARS-CoV-2 variant surges to prepare for immune escape and increased transmissibility.
Why it matters
Early warning of variant-driven case surges can guide pandemic response and save lives.
What the AI did
Strainflow used unsupervised latent-space features of SARS-CoV-2 genome sequences in a supervised model, trained on 0.9 million sequences, to predict case surges.
Results
Captured Delta and Omicron case rises two months ahead in most countries; predicted India surge as early as November 2021; prospectively validated July–December 2021.
Impact
Prospectively predicted variant-driven surges two months ahead, supporting pandemic preparedness.
Record
| Venue | Frontiers in Genetics |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Sargun Nagpal, Ridam Pal, Ashima Ashima, Ananya Tyagi, Sadhana Tripathi, Aditya Nagori, Saad Ahmad, Hara Prasad Mishra et al. |
| Citations | 35 (OpenAlex, at import time) |
| References | 51 |
| Model or method | Strainflow |
| Organization | not stated in source |
| Primary source | https://doi.org/10.3389/fgene.2022.858252 |
| Open access | gold |
| Keywords | COVID-19, genomic surveillance, language models, variant prediction, Strainflow |
Importance rating
major — millions of lives, an industry or a field
Pandemic-scale problem: global COVID-19 control threatened by emerging variants; early surge prediction aids preparedness across most countries, with prospective validation.