Digital innovation workshop "ART21", in response to constantly emerging new mutations of the coronavirus, undertook an ambitious project - to create a unique system for predicting the evolutionary development of the SARS-CoV-2 genome based on artificial intelligence and mathematical modeling technologies. In this way, the technology developed by Lithuanians will be able to predict the course and trends of the genome of this virus by using the data collected in the databases. The "learning" phase of the system has just ended, after which the results are encouraging - the accuracy is as high as 99%. If the results remain at least 80% after the second development of the technology, Lithuania will be able to be proud of having developed an advanced technology to fight the COVID-19 virus.
"The technology we are developing will make it possible to more effectively control the dynamics of the development of the pandemic. Based on the obtained results, test and vaccine developers will be able to make corrections in their production and avoid the loss of efficiency due to the appearance of new mutations in the viral genome. This is especially relevant in the test production process. For example, it may happen that a person will have the coronavirus and continue to carry it because the tests will fail to detect it and give false negative results. Therefore, this technology based on artificial intelligence and big data is especially relevant until the virus is fully contained. In addition, this is an excellent example of enabling open data to create advanced solutions even in such complex processes as disease prevention," said Augustas Alešiūnas, founder and director of the company developing the innovation, ART21.
According to the representatives of the company, there are only a few other attempts to create a similar technology in the world, but in this race they are far behind the Lithuanians. The uniqueness and complexity of the score prediction generator consists of several aspects.
Pharmaceutical companies respond to changes in real time by conducting research and noticing changes, so there is a time gap before corrections are made in the production of tests or vaccines. In this case, the creators of the innovation being developed are preparing to provide reasonable statistical forecasts of the future. Therefore, it will be possible to prepare for future mutations in advance. Another challenging factor that keeps other developers from taking on the cost risks of such projects is the incredibly long and dense data sequences. One line of such a sequence contains about 30 thousand. different signs, and there are tens of thousands of such lines. This not only requires specialists with special competence, but is also very time-consuming.
"Based on existing databases of the COVID-19 genome, we are building a dataset of mutation pairs. Genome sequences are compared, cleaned, various phylogenetic trees of the virus are constructed, we examine the entire genome of COVID-19 and the most important S (spike) protein separately. The obtained data of mutation pairs indicate the (possible) direction of mutation of the "parent" - "child" virus and allow training of a complex neural network - the mutation generator. We test the resulting model using ever-newly emerging virus mutations and hope in the future to reliably predict mutations into the future. The first results are promising and positive, but there is still a long, financial and time-consuming way to fully train, test and validate the system," one of the data scientists working with the technology development kitchen, Dr. Valdas Rapševičius.
