usDSM: a novel method for deleterious synonymous mutation prediction using undersampling scheme.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Synonymous mutations, while not changing amino acids, can impact RNA splicing and protein function.
• Advancements in sequencing highlight the importance of studying synonymous mutations in precision medicine.
• Current prediction methods for deleterious synonymous mutations are limited by imbalanced datasets.

Purpose of the Study:

• To address the challenge of imbalanced datasets in predicting deleterious synonymous mutations.
• To develop and validate a novel computational model for accurate prediction of deleterious synonymous mutations.
• To evaluate the role of deep learning in predicting deleterious synonymous mutations.

Main Methods:

• Expanded sample sizes from diverse data sources.
• Compared six undersampling strategies, identifying cluster centroid as most effective.
• Developed the undersampling scheme based method for deleterious synonymous mutation (usDSM) prediction using 14 biological features and a random forest classifier.

Main Results:

• Cluster centroid undersampling significantly improved dataset balance.
• The usDSM model demonstrated superior performance compared to state-of-the-art machine learning methods on test datasets.
• Deep learning models did not provide substantial benefits for deleterious synonymous mutation prediction in this context.

Conclusions:

• The developed usDSM model offers a more accurate approach to predicting deleterious synonymous mutations.
• The findings contribute to advancing computational methods for understanding human synonymous mutations.
• The usDSM web server is available for public use, facilitating further research.