AI-generated from publicly available materials.Anthropic's AI model, Claude Science, has demonstrated remarkable capabilities in protein design, outperforming human experts in a recent challenge.
In a recent evaluation, Claude Science successfully designed protein binders for 14 out of 15 specified targets, achieving superior results compared to previous design competitions. The AI’s performance was validated through collaboration with Adaptyc Bio and Twist Bioscience, where it utilized advanced models to achieve hit rates significantly above the typical 10-15% seen in conventional protein design efforts. Notably, when targeting each protein individually, the hit rate soared to 35.1%, showcasing Claude's efficiency in this complex task.
The study focused on well-researched targets to facilitate comparison with existing human designs. Claude not only surpassed the hit rates of human contest participants but also produced binders with higher affinity than the best entries from these competitions. While the AI had access to contest entries, it was instructed to avoid using existing binders as a foundation, demonstrating its ability to innovate independently. Anthropic emphasized that this foundational work is just the beginning, with plans to enable Claude to manage the entire drug development process across various modalities.
In addition to its protein design capabilities, Claude Science also showcased its potential to enhance drug development efficiency by expediting chemical analysis. The AI successfully processed nuclear magnetic resonance (NMR) and liquid chromatography-mass spectrometry (LC-MS) results in a fraction of the time typically required, producing results that closely matched those of traditional lab processing. This advancement could significantly streamline workflows in drug synthesis, highlighting the transformative potential of AI in life sciences.