The field of bio-computing has seen a major open-source contribution. ByteDance recently officially released a biological molecular structure prediction model named Protenix-v1. This model not only fully replicates the core capabilities of AlphaFold3 (AF3), but also announced that it will fully open-source the code and model parameters under the Apache 2.0 license, breaking through the technical barriers of top-tier biological large models.
The strength of Protenix-v1 lies in its full-atom 3D structure prediction capability, which can accurately handle complex biological systems including proteins, nucleic acids (DNA and RNA), and small molecule ligands. According to AIbase, this is the first fully open-source model that can match or even exceed AlphaFold3 in performance on multiple benchmark tests, under the same training data, model size, and inference budget.
To ensure fair and transparent evaluation, ByteDance simultaneously launched PXMeter v1.0.0 evaluation toolkit, which includes over 6,000 complex molecular samples, providing the industry with a reproducible standard benchmark. In addition, the project also offers a browser-based Web server, making it convenient for researchers to conduct interactive research directly. AIbase believes that the open-sourcing of Protenix-v1 will greatly accelerate innovation in cutting-edge fields such as drug development and synthetic biology.
