Key Findings
LPT successfully executed three end-to-end design campaigns initiated from single natural language prompts to disrupt CALCRL/RAMP1, PD-1/PD-L1, and PVR-TIGIT interactions. The framework autonomously identified design intent, selected targets, structures, and hotspot residues to design candidate binders. Cross-validation using multiple folding models indicates a high in silico success rate while maintaining low computational costs and minimal token consumption.
Technical Details
The LPT framework employs a fixed orchestration of LLM stages, where each stage operates within a bounded agentic loop featuring custom system prompts and specific tool access. To overcome the limitations of broad or outdated LLM training data, LPT utilizes a curated literature database containing approximately 14,000 papers digested into key findings. This specialized knowledge retrieval significantly improves accuracy: in tests using questions drawn from the database, the LLM with database access answered 41/48 questions correctly, whereas the model without access answered none.
Background & Context
Current large language models possess the ability to reason over scientific data and utilize code-based tools, but they often lack specialized knowledge and struggle with source attribution. Furthermore, unconstrained agentic workflows and unlimited tool use frequently result in the consumption of excessive amounts of tokens. LPT addresses these inefficiencies by bounding the agentic loops and providing a structured orchestration of stages.
Significance & Outlook
By combining bounded agentic loops with a specialized knowledge base, LPT enables the transition from a single natural language prompt to a completed protein binder design. This approach reduces the reliance on the broad, general knowledge of LLMs and minimizes the computational overhead typically associated with autonomous agentic workflows in scientific design.
Source: https://www.biorxiv.org/content/10.64898/2026.10.29.762730v1
Get our weekly technology intelligence — free
Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.
Subscribe Free — Weekly Tech Intelligence
By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.
- Your email and selected fields are used only to deliver the newsletter.
- We never share your information with third parties.
- You can unsubscribe anytime via the link in each email.
See our Privacy Policy for details.
Takes about a minute · Unsubscribe anytime
