Through the combined molecules, they were then fed into the virtual cell screening platform, where an enhanced version of Chu Xiaoxiao's AI model predicted their immune activation effects and potential cytotoxicity.
Ultimately, only a few dozen molecules reached the "final round."
At this moment, the reinforcement learning AI of the data hub began to come into play. It analyzed the common structural features, physicochemical properties, and performance data in previous modules of these dozens of successful molecules, summarizing a set of "success patterns." Then, it fed these patterns back to the initial design engine.
The second round of AI design was no longer a blind search starting from scratch but a targeted optimization based on the "successful experiences" of the first round! The newly generated molecules started from a higher baseline.
