Identification of Orphan Genes in Unbalanced Datasets Based on Ensemble Learning

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Orphan genes, crucial for regulatory patterns, are difficult to identify experimentally.
• Accurate classification of orphan genes in imbalanced datasets presents a significant computational challenge.

Purpose of the Study:

• To develop and evaluate a robust classification model for identifying orphan and non-orphan genes.
• To address the challenge of imbalanced datasets in orphan gene identification.

Main Methods:

• Synthetic Minority Over-sampling Technique (SMOTE) algorithms were employed to balance gene datasets.
• Ensemble classification algorithms, including Support Vector Machine, Random Forest, AdaBoost, GBDT, and XGBoost, were tested.
• The SMOTE-ENN-XGBoost model, combining over- and under-sampling with XGBoost, was proposed and evaluated.

Main Results:

• SMOTE combined with XGBoost achieved an F1 score of 0.94 on balanced Arabidopsis thaliana gene datasets.
• The SMOTE-ENN-XGBoost model demonstrated superior predictive accuracy compared to other SMOTE variants with XGBoost on imbalanced datasets.
• The proposed model provides a theoretical framework for orphan gene identification criteria.

Conclusions:

• The SMOTE-ENN-XGBoost model offers an effective solution for identifying orphan genes in imbalanced biological datasets.
• This approach provides a robust and computationally efficient method for advancing genomic research.