Deleterious synonymous mutation identification based on selective ensemble strategy.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Synonymous mutations are known to contribute to human diseases, but differentiating between deleterious and benign ones remains a significant challenge in medical genomics.
• Existing computational tools for predicting synonymous mutation effects often suffer from performance limitations due to imbalanced training datasets and insufficient negative samples.

Purpose of the Study:

• To develop a novel computational model, selective ensemble for predicting deleterious synonymous mutations (seDSM), to accurately identify deleterious synonymous mutations.
• To address the performance deficiencies of current methods by incorporating selective ensemble strategies and handling imbalanced data.

Main Methods:

• Constructed base classifiers using balanced training subsets from imbalanced datasets.
• Employed pairwise diversity metrics to select diverse classifiers for integration via soft voting.
• Investigated strategies for missing value imputation and model construction using different diversity metrics, identifying double fault with EKNNI as optimal.
• Developed the seDSM model using 40-dimensional biological features.

Main Results:

• The selective ensemble approach, particularly using double fault diversity and the EKNNI imputation strategy, demonstrated superior performance.
• The proposed seDSM model significantly outperformed existing state-of-the-art methods on independent test sets based on multiple evaluation metrics.
• seDSM exhibits outstanding predictive performance for deleterious synonymous mutations.

Conclusions:

• The developed seDSM model offers a powerful and accurate tool for predicting deleterious synonymous mutations.
• seDSM has the potential to advance the study of synonymous mutations and their role in human diseases.
• The source code for seDSM is publicly available, facilitating further research and application.