The wily SARS-CoV-2 has mutated to the point where current vaccines may need to be revised—via advanced pattern recognition AI.

A team of researchers at the University of Southern California (USC) Viterbi School of Engineering has developed a unique method employing AI to expedite the analysis of COVID-19 vaccines as well as deduce the best possible preventive medical therapy for containing the current coronavirus pandemic.

The method is easily adaptable to analyze ongoing mutations of the virus, to help governments identify to most suitable vaccine candidates, according to the research report. The machine learning (ML) model is said to accomplish vaccine design cycles, which used to take months or years, in a matter of seconds.

Paul Bogdan, Associate Professor of Electrical and Computer Engineering and corresponding author of the study by USC Viterbi, was quoted saying: “This AI framework, applied to the specifics of this virus, can provide vaccine candidates within seconds and move them to clinical trials quickly to achieve preventive medical therapies without compromising safety. Moreover, this can be adapted to help us stay ahead of the novel coronavirus as it mutates around the world.”

NLP and viral mutations

A transmission electron micrograph of a B.1.1.7 variant coronavirus, whose increased infectiousness is believed to be due to changes in the structure of the spike proteins, shown here in green. Source: Wikipedia

In addition to AI, one technology that could be used by researchers to decode the properties of biological systems is Natural Language Processing (NLP). According to the USC scientists, the mutation of protein sequences and genetic codes can be modeled using NLP techniques.

Just as good grammar and semantics can improve communication, the genetic code of a mutating virus needs to observe the rules of biochemistry to remain viable and increase survivability in the resulting variant. NLP models work by encoding words in a mathematical space in such a way that words with similar meanings are closer together than words with different meanings. This is known as embedding. When applied to how viral code mutates, NLP can group genetic sequences according to how similar certain patterns of mutations are. This in turn helps researchers to improve their chances of identifying vaccine candidates that can target specific groups of mutation patterns.

Patterns can also be discerned via visual clues. According to Dr Hemang Shah, who serves as the India Engineering Lead of Qualcomm, AI-based visual processing algorithms can predict patterns with high accuracy. “There was a recent project on diabetic retinopathy that involved a vision-based algorithm, which could detect malfunctions based on patterns. It also worked in tandem with employing X-ray images and scans of the lungs and other internal organs.”

Satya Mallick, Chief Executive Officer of, a platform that provides a real-time optimized computer vision library, tools and hardware, affirmed that all patterns could be analyzed once there is enough data. “AI can do better than what humans are capable of in pattern recognition. It is certainly in the realm of possibility,” he said.

Finally, according to Nikhil Bhaskaran, Founder & Chief Executive Officer, ShunyaOS, a low code AIoT embedded platform: “Last year, we undertook a similar research-based project with a few hospitals in India involving ML and deep neural network algorithms to help predict patterns of viral mutations. As with the USC project, we should also do an AI-powered study on the lack of Oxygen in our country by collecting healthcare data at the ground level.”

26 potential vaccines identified?

Virus mutations are a common phenomenon, and in the case of SARS-CoV-2, scientists and virologists have so far identified the Brazil variant, the UK variant, the South Africa variant and quite recently, the B.1.167 ‘double mutant’ due to the presence of two mutations in the virus. Researchers are alluding to the possibility that this variant is even more infectious than the UK variant.

When it comes to treatment therapies, USC’s computer model has already eliminated 95% of the numerous compounds considered in alleviating COVID-19 symptoms, pinpointing the most viable remaining candidates, according to the study report.

Also, the USC research team has asserted that their system has predicted 26 potential vaccines that could work against the coronavirus. From those, it has identified the best 11 from which to construct a ‘multi-epitope’ vaccine that can attack the ‘spike protein’ of the coronavirus. This refers to the part of the virus to binds to cell surfaces and subsequently opens a gate through which the virus can enter the cells.

The right vaccine candidates can disrupt the different spike proteins of various mutations, thereby neutralizing the ability of the any mutated virus to enter cells and replicate.