ML Document Annotation for Biomedical Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current databases like PubMed lack intelligent access to combinations of biomedical information, particularly for functional studies, and require institutional licenses or payments, limiting access to over two million documents without a comprehensive way to retrieve specific information on entities like genes, drugs, and diseases.
Innovation Solution
A system using machine learning and natural language processing to classify and annotate documents into functional and clinical categories, allowing for intelligent access by identifying relationships between entities such as genes, gene variants, drugs, and diseases, and ranking documents based on relevance and frequency of search terms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If institutional license agreements are implemented with journal publishers, then access to research articles is granted, but cost and accessibility barriers are created for users without institutional licenses
Solution Approach 1:
The system creates and uses annotations as simplified copies or representations of the full research articles. These annotations capture key information about genes, gene variants, drugs, and diseases, allowing users to access essential biomedical information without needing to obtain or pay for the full licensed articles. The annotation module extracts and stores critical data points that replicate the informational value needed for research queries.
Solution Approach 2:
The annotation module extracts specific key information from research articles, separating essential biomedical data (about genes, gene variants, drugs, diseases) from the full article text. This extraction creates standalone annotation records that can be searched and accessed independently, removing the dependency on accessing full licensed articles while preserving the core informational content.
2Ease of operation
If a comprehensive classification system is implemented for all document types, then information organization is improved, but complexity of the classification system increases
Solution Approach 1:
The system segments the classification task into two distinct parts: (1) a robust existing classification system for clinical articles that PubMed already maintains, and (2) a new annotation module that specifically handles functional studies. This segmentation allows each part to be optimized independently, avoiding the need to redesign the entire classification system while still achieving comprehensive organization of all document types.
Solution Approach 2:
The annotation module serves multiple functions simultaneously: it classifies functional studies, extracts key biomedical information, creates searchable annotations, and integrates with existing PubMed search infrastructure. This multi-functionality reduces the need for separate systems for each task, simplifying the overall architecture while maintaining comprehensive classification capabilities.
3Measurement precision
If machine learning annotation modules are used to classify documents, then classification accuracy is improved, but computational resources and processing time are increased
Solution Approach 1:
The system performs preliminary annotation and classification of functional studies in advance, creating a pre-processed database of annotated records before users submit search queries. The annotation module analyzes and tags articles with key biomedical information during an initial processing phase, so that when users search, the results are immediately available without requiring real-time analysis. This preliminary action shifts processing time from query-time to offline batch processing.
Data Source
AI summary
Computer-based methods, systems, and computer readable media for managing documents within a content repository or documents within the document subsets are provided. Documents within the content repository may be classified into one of a functional category and a clinical category. Documents are applied to a machine learning annotation and analysis module to automatically annotate the documents to indicate relationships between entities. A request is processed for the documents including one or more search terms, wherein the search terms pertain to one or more entities from a group of gene, gene variant, drug, cancer and a biomedical/clinical term. Documents satisfying the request are identified by comparing the one or more search terms to the annotations and specific sections of the documents, and determining a relevance of a document based on the comparison and a frequency of the one or more search terms in each of the specific sections. The identified documents are ranked according to custom techniques.