The model can detect around 73% of variants in each country leading to at least 1,000 cases per 10 lakh people within three months.In the face of a renewed surge in COVID-19 cases, a recent study reveals that an artificial intelligence (AI) model, developed by researchers from MIT in the US and The Hebrew University-Hadassah Medical School in Israel, can predict which variants of the SARS-CoV-2 virus are likely to lead to new waves of infection. The model shows the ability to detect about 73% of variants in each country that result in at least 1,000 cases per 10 lakh people in the three months following a one-week observation period, increasing to over 80% after two weeks.
The research team conducted a comprehensive analysis using 9 million genetic sequences of the SARS-CoV-2 virus from 30 countries. This genetic data was sourced from the Global Initiative on Sharing Avian Influenza Data (GISAID), an initiative that facilitates the swift sharing of data related to priority pathogens such as influenza and the coronavirus. The researchers combined this genetic information with data on vaccination rates, infection rates, and other relevant factors. By integrating diverse datasets, the study aimed to gain a more comprehensive understanding of the virus’s genetic variations and their correlation with various epidemiological factors.
Using the patterns found in their genetic sequence analysis, the researchers developed a machine learning-based risk assessment model. An AI algorithm that learns from historical data to generate predictions is called machine learning. The study’s conclusions, which were published in the journal PNAS Nexus, emphasized important variables affecting how contagious a variant is. These include the spike mutations it carries, the early trajectory of the infections it causes, and the differences in its mutations from the dominant variant during the observation period. The model seeks to support proactive public health initiatives by offering insights into the possible effects of various virus variants.
The researchers draw attention to a weakness in current models that makes it difficult to forecast how virus variants will spread. Their novel method improves early signals and forecasts the future transmission of recently discovered variants by combining genetic and epidemiological data specific to the variant. This gives rise to a more comprehensive tool for managing and anticipating the changing dynamics of viral spread.
