CNV-ETLAI

CNV-ETLAI automates extraction, transformation, and organization of copy number variation (CNV) information from scientific literature into a structured database to support detection and interpretation of CNVs.

Key Features:

• Automated Data Extraction: Integrates text, table, and image processing algorithms to identify and extract CNV information from journal articles.
• Data Transformation and Organization: Transforms raw CNV data into a structured database format for downstream analysis and querying.
• Artificial Intelligence Analysis: Leverages advanced artificial intelligence algorithms to automate the extraction, transformation, and loading of CNV data.
• Validation Against Human Experts: Validation studies report a 4% improvement in CNV detection accuracy compared to human experts using ground truth datasets labeled by experts.
• Speed and Scalability: Performs analyses approximately 60 times faster than manual methods and is expected to improve performance as the database scales.
• Comprehensive Literature Coverage: Compiled data on 5,800 CNVs from 2,313 journal articles.
• Chromosome-Specific Insights: Analysis of collected data showed chromosome X had the highest total CNV frequency across its length while chromosome 22 had the highest CNV frequency per megabase.
• Crowdsourcing Analysis: Employed a crowdsourcing approach to further analyze the collected CNV data.

Scientific Applications:

• CNV classification and interpretation: Rapidly processes literature-derived CNV data to support classification and interpretation for research and clinical decision-making.
• Support for geneticists and clinicians: Provides a curated CNV database resource for geneticists, molecular biologists, and clinicians studying or diagnosing diseases associated with genomic variations.

Methodology:

Integrates text, table, and image processing algorithms and advanced artificial intelligence algorithms to perform extraction, transformation, and loading into a structured database, with validation against human-expert–labeled ground truth datasets and additional analysis via crowdsourcing.