Protenix’s open-source reimplementation of AlphaFold3 is indeed powerful—but the core backbone took the technology and left to start a business. Then AML, after coming under Yang Zhen’s command, still wasn’t proactive. This adjustment to the script is a bit subtle: the drug discovery pipeline is too long—from nanomolar hit rates to IND validation—where capital’s patience is harder to withstand than the technology.

According to Beating, ByteDance's Seed AI for Science team is discussing a new round of adjustments, planning to transfer from AML to Yang Zhenyuan's leadership, but the other party is not enthusiastic. Core key members Gu Quanquan, Xiao Wenzhi, and others have already left to start their own businesses, focusing on AI drug and protein design, and have received USD investments. The team's achievements include Protenix (an open-source reproduction of AlphaFold3) and PXDesign, achieving a 20%–73% nanomolar hit rate across six targets and five points. The drug discovery chain is long, requiring wet lab validation, animal testing, IND, and BD verification to promote asset development.