Last Updated on 2026-02-20
In the Kaggle competition “Stanford RNA 3D Folding,” our software engineer Oda placed 2nd out of 1,414 teams, winning a gold medal and prize money.
“Stanford RNA 3D Folding” is a global machine learning competition aimed at predicting the three-dimensional structures of RNA.
RNA 3D structure prediction is considered one of the “greatest remaining challenges” in biology. This highly challenging and significant endeavor aims to create an RNA version of “AlphaFold,” which revolutionized protein 3D structure prediction.
https://www.kaggle.com/competitions/stanford-rna-3d-folding
Oda adopted the “RBSSA (Representation-Based Sequence Search and Alignment)” method, which utilizes the internal representation of "RibonanzaNet https://www.biorxiv.org/content/10.1101/2024.02.24.581671v2,“ an RNA-based model. This approach incorporates cutting-edge insights from Oda's field of expertise, protein science, which states that ”the internal representations acquired by Deep Learning models capture the ‘structural similarity’ of biomolecules extremely well."
Subsequent analysis after the competition showed that RBSSA tends to yield different hits compared to conventional sequence search, while maintaining comparable overall accuracy. This complementary effect with conventional sequence search likely contributed to the high scores achieved.
For detailed solution methodology, please see:
https://www.kaggle.com/competitions/stanford-rna-3d-folding/writeups/2nd-place-solution
https://www.youtube.com/watch?v=kKC5WjAqWsc
QuettaAI, in collaboration with its technology partner PEZY Computing, leverages its experience in developing “medical imaging diagnostic devices” and “genome analysis systems” to strongly promote the utilization of generative AI in healthcare fields such as drug discovery.