usDSM

usDSM predicts deleterious synonymous mutations that can affect RNA splicing regulation and transcript processing to support interpretation of their functional consequences in precision medicine.

Key Features:

• Imbalanced dataset handling: Employs undersampling strategies to balance positive (deleterious) and negative (neutral) samples in training data.
• Undersampling strategy comparison: Evaluates six different undersampling strategies and identifies the cluster centroid method as the most effective.
• Feature representation: Integrates a 14-dimensional set of biological features for mutation representation.
• Classifier: Uses a random forest classifier to predict deleterious synonymous mutations.
• Performance evaluation: Demonstrates superior performance compared to other state-of-the-art machine learning methods on various datasets, with deep learning models not showing substantial advantages for this task.

Scientific Applications:

• Precision medicine: Improves prediction of deleterious effects of synonymous mutations to inform precision medicine research.
• Interpretation of genetic variation: Aids understanding of how synonymous variants impact RNA splicing and the functional consequences of genetic variation, supporting diagnostic and therapeutic investigations.

Methodology:

Evaluates six undersampling strategies and adopts the cluster centroid method, integrates 14-dimensional biological features, applies a random forest classifier, and assesses performance across multiple datasets.