You are here

Baidu's mRNA vaccine optimization algorithm featured in Nature

Baidu and its collaborators made a significant breakthrough in biocomputing when they proposed the mRNA sequence optimization algorithm LinearDesign in the May issue of Nature, a leading international academic journal, which was titled "Algorithm for Optimized mRNA Design Improves Stability and Immunogenicity."

A Chinese IT firm has broken new ground by being the first to submit and have a manuscript accepted for publication in Nature. Since this is the first CNS major issue publication on the subject of AI applied to mRNA, Nature posted the preview version of the fast track (expedited article preview) online before the official layout.

Although mRNA vaccines have been identified as a potentially useful method for controlling COVID-19, they still must overcome a number of obstacles. One of the challenges in mRNA therapy research is figuring out how to efficiently build stable and improved drug-forming mRNA sequences.

Immunizing patients with mRNA vaccines expressing tumor-associated antigens (TAAs) is the most straightforward application of mRNA vaccines in oncologic settings. The second way mRNA vaccines are being put to use is in the development of targeted cancer vaccines. In cellular therapeutics, mRNAs are used to transfect patient-derived cells in vitro before reinfusing the altered cells back into the patient.

mRNA vaccine design encompasses antigen identification and selection, mRNA format determination, mRNA modification, and delivery route and formulation determination.

To enhance vaccine stability and efficacy, Baidu applies Lattice Parsing technology from the field of natural language processing to mRNA vaccine sequences.

In the case of the SARS-CoV-2 Spike protein, for example, to find a stable sequence using the traversal method, one would need to look at 10 to the power of 632 mRNA sequences. If a supercomputer were to calculate one sequence per second, it would not be able to calculate even one billionth of the potential mRNA sequences in the 13.8 billion years since the birth of the universe, whereas the LinearDesign algorithm can find the most stable candidate sequence in just 11 minutes.

LinearDesign, an algorithm to generate sequences, has been shown through experiments to aid biopharmaceutical companies speed up the production of more effective mRNA vaccines while cutting down on development efforts and costs. The COVID-19 mRNA vaccine and the Herpes zoster mRNA vaccine are two examples of how successful this method could potentially be. The antibody response in vivo can be improved by as much as 128-fold thanks to Baidu's design compared to conventional gold standards, which primarily focus on in vitro mRNA half-life and protein expression.

"The new approach is remarkable," according to computational RNA biologist Dave Mauger, who worked at Moderna, an American manufacturer of mRNA vaccines. "The computational efficiency is really impressive, and it's more sophisticated than ever before."

To date, Baidu has developed PaddleHelix, a comprehensive flying paddle-based biocomputing platform that encompasses Wenxin large models–biocomputing big models and investigates the use of artificial intelligence in small molecule, protein/peptide, RNA, and other settings.