$\texttt{PatentAgent}$: Intelligent Agent for Automated Pharmaceutical Patent Analysis

TL;DR

This work introduces the first intelligent agent in this domain, poised to advance and potentially revolutionize the landscape of pharmaceutical research, which comprises three key end-to-end modules that perform patent question-answering, image-to-molecular-structure conversion, and core chemical structure identification.

Abstract

Pharmaceutical patents play a vital role in biochemical industries, especially in drug discovery, providing researchers with unique early access to data, experimental results, and research insights. With the advancement of machine learning, patent analysis has evolved from manual labor to tasks assisted by automatic tools. However, there still lacks an unified agent that assists every aspect of patent analysis, from patent reading to core chemical identification. Leveraging the capabilities of Large Language Models (LLMs) to understand requests and follow instructions, we introduce the $\textbf{first}$ intelligent agent in this domain, $\texttt{PatentAgent}$, poised to advance and potentially revolutionize the landscape of pharmaceutical research. $\texttt{PatentAgent}$ comprises three key end-to-end modules -- $\textit{PA-QA}$, $\textit{PA-Img2Mol}$, and $\textit{PA-CoreId}$ -- that respectively perform (1) patent question-answering, (2) image-to-molecular-structure conversion, and (3) core chemical structure identification, addressing the essential needs of scientists and practitioners in pharmaceutical patent analysis. Each module of $\texttt{PatentAgent}$ demonstrates significant effectiveness with the updated algorithm and the synergistic design of $\texttt{PatentAgent}$ framework. $\textit{PA-Img2Mol}$ outperforms existing methods across CLEF, JPO, UOB, and USPTO patent benchmarks with an accuracy gain between 2.46% and 8.37% while $\textit{PA-CoreId}$ realizes accuracy improvement ranging from 7.15% to 7.62% on PatentNetML benchmark. Our code and dataset will be publicly available.

Figures (6)

• Figure 1: Comparison of patent analysis between manual, traditional machine learning, and PatentAgent. The sub-tasks listed in the Traditional Machine Learning phase of patent analysis are: Named Entity Recognition (NER), Object Detection (OD), Optical Chemical Structure Recognition (OCSR), and Core Chemical Structure Identification (CCSI).
• Figure 2: Workflow Illustration for PatentAgent. PatentAgent consists of three major modules, namely PA-QA, PA-Img2Mol, and PA-CoreId. PA-QA processes user requests and outputs patent segmentation (texts or images), PA-Img2Mol processes molecular images and output SMILES, and PA-CoreId processes SMILES and outputs the core chemical structure. Refer to the color coding legends for better understanding.
• Figure 3: Complete workflow of converting a molecular image to SMILES: from model inference through VLM evaluation to final output.
• Figure 4: This figure shows the process of extracting related images of chemical structure in pdf patent file.
• Figure 5: The left image illustrates the framework for achieving specific lead identification (leadid), covering the process from feature selection to model training. The right image shows an example of leadid classification within a patent: (a) in the top left visualizes the lead id, (b) in the bottom left projects the molecules from the patent and marks the position of the lead id and (d) in the top right displays a probability distribution graph of the molecules in the patent.