Patent LLM

Overview

The Patent LLM system helps translational research teams assess druggability and competitive landscape around specific gene targets by:

• Target Validation Acceleration: Quickly determining if a target is druggable by finding patents that have already attempted to target the gene of interest

• Competitive Intelligence: Identifying where a target falls in the landscape ("sweet spot" with few patents, "no patent" higher risk, or "many patents" requiring strategy pivot)

• Research Efficiency: Automating patent searches that would otherwise require significant manual effort

• Informed Decision Making: Supporting decisions about which targets to pursue, whether to build in-house or seek in-licensing, and competitive positioning

Output formats

The Patent LLM agent returns natural language summaries with structured elements embedded in conversational text. The output is designed to provide both high-level overview and detailed patent-by-patent analysis, with references and links embedded within the narrative.

Query examples by use case

IP landscape analysis

Purpose: Map the patent landscape around a target or mechanism to understand who owns what IP and identify crowded vs. open spaces.

Example Prompt:

"I'm interested in DGAT2 as a therapeutic target. Are there patents related to this target? In which disease areas? Please cover as many modalities as possible (small molecules, antibodies, etc.), discard abandoned patents, and include a summary of the claims as well as the entity submitting the patent. Include correct links (no hallucinations).

Target druggability assessment

Purpose: Determine if a target has been successfully pursued before (evidence of druggability) and understand what classes of molecules have been tested.

Example Prompt:

"Are there patents covering assets against target CD73? I want to understand if this target is druggable and already have small molecules or antibodies been developed against it?"

Freedom-to-operate research

Purpose: Assess whether developing a therapeutic against a specific target will infringe existing patents; identify potential blocking patents.

Example Prompt:

"I'm developing a small molecule inhibitor targeting KRAS G12C. What US patents exist that cover KRAS G12C inhibitors? Please identify the key claims and the companies holding these patents. Highlight any patents that are still active (not expired or abandoned)."

Competitive intelligence

Purpose: Understand the competitive landscape—who is working on the same target, what development stage, and what's the strategic positioning.

Example Prompt:

"What is the competitive landscape for PD-L1 as a therapeutic target? Which companies hold patents? What are the key mechanistic approaches (checkpoint blockade, antibody-drug conjugates, bispecifics)?"

Early research (target ID and prioritization)

Purpose: During target discovery/validation, use patent data as one signal to prioritize targets—understanding which have precedent and which are novel.

Example Prompt:

"I have a list of candidate targets for ovarian cancer: FOLR1, NECTIN4, TROP2, and CLDN6. For each target, identify how many US patents exist that claim therapeutic assets. I want to understand which targets have the most precedent (druggability signal) vs. which are under-explored."

In-licensing (asset assessment)

Purpose: During due diligence for in-licensing or acquisition, assess the IP landscape around the asset's target and mechanism to understand competitive risk and differentiation.

Example Prompt:

"I'm assessing a NECTIN4-targeted ADC for in-licensing. Are there existing patents covering NECTIN4 ADCs? What do the claims cover—target specificity, linker chemistry, payload, or combination strategies?"

Limitations

• Currently, the tool only leverages US patents (USPTO via PatentsView).