The portfolio of North American graphics chip manufacturer, Nvidia, is once again in the process of expanding due to the development of generative artificial intelligence. Through a pioneering application of new technology, the manufacturer created a language model capable of analyzing data from the SARS-CoV-2 virus responsible for Covid-19 and, based on this, predicting possible mutations.
Development of GenSLMs
The new language model (LLM), called GenSLMs, was created in a joint effort by the chipmaker with the University of Chicago and Argonne National Laboratory. Like all GenAI technology, the new model was trained using information relevant to its operation — in this case, the genomic data of SARS-Cov-2. More specifically, around 110 million genomes were used during the training period, allowing GenSLMs to distinguish, classify and group relevant DNA sequences.
Through its impressive ability to distinguish these nucleotide sequences that form the basis of DNA and RNA, GenSLMs has a deep understanding of the relationship between the virus and its variations, allowing it to predict future derivations of SARS-Cov-2.
The 3D model of the strains derived by GenSLMs, based on genomic information from ancient SARS-Cov-2 variations (Photo: reproduction/Nvidia/Argonne)
According to information provided in an official statement from Nvidia, the training appears to have had the desired effect. Even though it was only fed with information from the original virus and its respective Alpha and Beta variations, GenSLMs was able to analyze and deduce the existence of other genetic mutations present in current strains of Covid-19.
And as Arvind Ramanathan, the project’s Chief Investigator and Argonne affiliate, wrote, this is a result that “convincingly validates the capabilities of the new language model”.
Impact of AIs in the field of Genetics
The success of the new language model focused on Covid-19 analysis and prevention is not alone in medicine, sharing the spotlight with other LLMs such as Ankh and CancerGPT. In a technological era of constant development, the presence of generative AI is beginning to significantly advance the progress of modern genetic research.
The biggest differentiator of this technology is the ability to analyze and learn from a set of specific data to be able to consistently predict and generate new data that is relevant to researchers. In the field of genetics, for example, this means being able to work with complex genomic sequences.
A GenAI called Ankh, for example, was developed in a collaborative effort between the Universities of Munich and Columbia to analyze the deep language of proteins. Meanwhile, CancerGPT was created to predict outcomes from drug-based cancer treatments at the Universities of Texas and Massachusetts.
Regardless of their respective focuses, the use of generative AI technology in these studies signals an important paradigm shift in information processing and derivation.
Featured photo: Nvidia headquarters in Santa Clara, Silicon Valley, California. (Reproduction/Bloomberg/Philip Pacheco).