A team of researchers from the Institut de Biologie Paris Seine (IBPS) at Sorbonne University in collaboration with the University of Lausanne, has used machine learning techniques to predict the probability of new mutations appearing in SARS-CoV-2, the virus that causes Covid-19.
The Covid-19 pandemic has been characterised by the repeated emergence of new variants, the latest being Omicron.
Each of these variants have distinguishing mutation patterns that may enable it to spread quickly or reduce the vaccine’s effectiveness.
Mutation leads to changes in surface proteins. Amino acids are the building blocks of proteins. Sites refer to the positions of amino acids in the spike protein. The spike protein is located on the outside of a coronavirus and is how SARS-CoV-2 (the coronavirus) enters human cells.
For example, as a result of new mutations, the Omicron variant is spreading quicker than the earlier variants and even manages to partially escape the immune response to prior infection.
Given the impact of the virus at the start of the pandemic in 2020, predicting as yet unseen mutations could help in mitigating the impact of the pandemic.
Using a machine learning technique called generative modeling, the team led by Professor Martin Weigt of the Computational and Quantitative Biology lab at IBPS used a single Sars-Cov-2 genome sequence of the original Wuhan strain, in combination with the publicly available sequence data of other coronaviruses, to predict the possible mutation sites on the Covid-19 virus.
“We used the accumulating sequence diversity of the coronaviruses during the Covid-19 pandemic to validate our method and to check our prediction with what is actually happening,” Weigt told RFI.
Combining the results of the computational modeling with immunological data allowed the researchers to identify positions in the receptor binding domain of the virus’ spike protein (which plays a central role in its transmission in humans) which are susceptible to mutations and play an important role in the human immune response to a Covid-19 infection.
“As per our predictions, we identified nine positions, which are mutable and immunogenic. Four of them were already mutated in earlier variants of concern. Delta carried one single mutation in these sites, Omicron carries four, one of them being new as compared to earlier variants of concern.”
“We find that four out of the 14 mutations in Omicron's receptor binding domain are in these nine "dangerous" sites out of 178 possible sites, which is a strong enrichment.”
Weigt said the virus will continue to mutate as long as it is circulating among us.
He added that a combination of these nine positions could give rise to future mutations that have an even increased potential for immune escape, “meaning they could be nasty variants capable of escaping the immune response.”
Weigt stressed that their approach was probabilistic. “What we predict is mutability, or the potential to undergo mutations. Statistical modelling tells us that certain mutations have a higher probability to appear than others. We cannot predict if and when they will appear,” he said.
He added that the distinguishing feature of this study is the use of computational modelling for viruses where data is quite limited. “We pushed the modelling techniques to the limits of their applicability,” he said.
“Typically, we love good data. That’s why we have been using this method quite a lot in bacteria. [We used this technique] for example, for studying and predicting the evolution of proteins of interest like the enzyme that digests penicillin and makes the bacteria antibiotic drug resistant.”
Weigt said he and his colleagues were also astonished to find how well the computational technique worked for viruses for which the data isn’t as substantial as is the case with bacteria.
