Identification and prioritization of disease candidate genes using biomedical named entity recognition and random...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Identifying candidate genes is crucial for understanding complex diseases, enabling early diagnosis, personalized treatment, and drug discovery.
• Traditional methods for disease gene identification face limitations, requiring large sample sizes and high statistical power, which is challenging for complex diseases.
• Existing disease gene prioritization methods often rely on computational predictions lacking experimental validation.

Purpose of the Study:

• To introduce an innovative two-tier machine learning protocol for distilling disease-gene association details from disease-specific abstracts.
• To classify disease-gene associations into Positive, Negative, and Ambiguous categories using advanced text mining.
• To address the limitations of existing tools in disease gene identification and prioritization.

Main Methods:

• Utilized a two-tier machine learning protocol employing Random Forest for text classification.
• Level 1 classifier distinguished the presence or absence of disease-gene associations.
• Level 2 classifier further categorized associations into Positive, Negative, and Ambiguous classes, trained and validated on Alzheimer's, Breast Cancer, and Type 2 Diabetes datasets.

Main Results:

• Achieved high accuracy: 97.29% for level 1 and 98.14% for level 2 classification.
• Successfully extracted 2769, 3220, and 740 positively associated genes for Alzheimer's, Breast Cancer, and Type 2 Diabetes, respectively.
• Identified a significant number of novel genes (1008 for Alzheimer's, 670 for Breast Cancer, 165 for Type 2 Diabetes) not present in existing databases, expanding genetic exploration.

Conclusions:

• The developed protocol offers a comprehensive roadmap for exploring genetic diseases.
• The protocol demonstrated versatility and robustness across different disease contexts (Alzheimer's, Breast Cancer, Type 2 Diabetes).
• The findings provide promising candidates for targeted interventions and highlight ambiguous associations for further investigation, with broad applicability in biomedical research.