Genesis Molecular AI's CTO Sergey Edunov clarifies what Claude actually did in Anthropic's protein binder work: Claude was the orchestrator, not the discoverer. The heavy lifting—finding the actual binders—came from specialized scientific models underneath.
Key technical points:
• Claude's role = coordination layer for molecular tools, not the core discovery engine
• The prompt fed to Claude is 16,000 words long (essentially a mini technical manual)
• Important limitation: protein binders ≠ drugs. They're precursor components, not therapeutic modalities you can directly use in patients
This is a critical distinction for understanding AI in drug discovery. LLMs like Claude can route tasks and parse complex scientific instructions, but domain-specific models (trained on molecular data) do the actual computational biology work. Think of Claude as the project manager reading a massive spec doc, while the specialized models are the engineers building the thing.
Key technical points:
• Claude's role = coordination layer for molecular tools, not the core discovery engine
• The prompt fed to Claude is 16,000 words long (essentially a mini technical manual)
• Important limitation: protein binders ≠ drugs. They're precursor components, not therapeutic modalities you can directly use in patients
This is a critical distinction for understanding AI in drug discovery. LLMs like Claude can route tasks and parse complex scientific instructions, but domain-specific models (trained on molecular data) do the actual computational biology work. Think of Claude as the project manager reading a massive spec doc, while the specialized models are the engineers building the thing.