A high-performance approach for predicting donor splice sites based on short window size and imbalanced large samples

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Splice site prediction is crucial for accurate gene structure determination.
• Existing computational methods face challenges with large, imbalanced datasets and optimal window size selection.
• Short window sizes can improve predictive accuracy and reduce computational cost.

Purpose of the Study:

• To develop a novel computational method for donor splice site prediction.
• To address challenges posed by imbalanced sample sizes and optimize window size selection.
• To enhance the accuracy and efficiency of splice site prediction in bioinformatics.

Main Methods:

• Developed the chi-square decision table (χ²-DT) method utilizing a short window size (11 bp).
• Employed chi-square tests for feature extraction and information gain for feature selection.
• Constructed a balanced decision table to handle imbalanced classification with a large training set (2000 true sites: 271,132 false sites).

Main Results:

• χ²-DT achieved the highest independent test accuracy (93.34%) compared to other classifiers.
• Demonstrated high accuracy (92.40%) on mutated sequences with frameshift errors.
• Outperformed both long and short window size-based methods in predictive accuracy and computational speed (89s).

Conclusions:

• The proposed χ²-DT method offers superior predictive accuracy for donor splice sites.
• Achieves high computational speed and robustness against insertions/deletions.
• Provides an effective solution for splice site prediction using imbalanced large samples and short window sizes.