Felix Pinkston
April 11, 2025 15:42
Stanford’s DAS Lab develops RNA folding research using NVIDIA DGX Cloud to develop more accurate RNA models using community participation and state -of -the -art technology.
DAS Lab of Stanford University is a significant progress in RNA folding research by utilizing advanced computing features of NVIDIA DGX Cloud. This initiative, supported by the NAIRR pilot program, can access 32 NVIDIA A100 DGX cloud nodes with eight GPUs in the laboratory for three months. According to NVIDIA, this practical computational ability has made the laboratory to facilitate the training of large -scale models and data sets by switching from small experiments to large -scale distributed education.
Community -oriented research
Under Dr. Rhiju Das’s leadership, DAS Lab was at the forefront of RNA research. In 2020, the laboratory hosted the Openvaccine Kaggle competition in response to Covid-19 Pandemic and continued to work with Ribonanna Competition in 2024. These initiatives aim to accelerate understanding of RNA structure and biological functions.
One of the main obstacles to RNA folding studies is the lack of experimental RNA structure data. To overcome this, DAS Lab developed Eterna, a crowdsourcing game that allows users to design RNA sequences. These sequences are synthesized in the laboratory, and chemical mapping experiments are performed to infer the RNA structure.
Innovative strategy
The laboratory strategy includes some innovative approaches.
- Collection of crowd sourcing data: Eterna is used to collect new RNA sequences from the public supplemented by a database in which an expert is built.
- Data approximation: The chemical mapping experiment creates a reactive profile that helps to approximate the RNA structure.
- Model design through crowd sourcing: The laboratory uses the KAGGLE competition to test various model architectures and test the training pipeline along with the community participation.
The laboratory also accelerated the creation of a new sequence by developing a trained reinforced learning model trained to play Eterna. This model used 4,000 A100 GPU time at NVIDIA DGX Cloud and was trained using Q-running algorithms.
Amazing results
DAS Lab successfully selected the largest database for RNA rescue education. The Foundation model trained with the 256 A100 GPU led to the development of ribo Nannet Net 2 and is currently achieving cutting -edge performance in RNA folding. This model is now available for community use and fine adjustment.
On February 26, 2025, the Institute began competition for Stanford RNA 3D Folding Kagle, providing $ 75,000 in prizes to encourage further improvement of Ribonanzanet2 to predict RNA structure. The competition invites participants to take advantage of the experimental RNA structure collected during the contest.
Future prospect
The research conducted by DAS Lab has significant potential for the development of biological science that affects medicine, agriculture and biotechnology. By developing a more accurate RNA model, researchers can better understand the disease mechanism and create more effective treatment.
In the future, DAS LAB plans to expand the data set and model using more powerful calculation resources provided by NVIDIA DGX Cloud. Their work shows the power of crowd sourcing and cutting -edge technology in developing science research.
Image Source: Shutter Stock