CarSite-II

CarSite-II predicts carbonylated sites within proteins to identify residues modified by reactive oxygen species (ROS), specifically targeting lysine (K), proline (P), arginine (R), and threonine (T).

Key Features:

• Target residues: Identifies carbonylated residues at lysine (K), proline (P), arginine (R), and threonine (T) resulting from ROS-induced carbonylation.
• Carbonylation biology: Accounts for carbonylation as an irreversible post-translational modification caused by reactive oxygen species that converts amino acid side chains into carbonyl products.
• Resampling strategy: Balances class proportions using SMOTE-KSU, a resampling approach that combines K-means similarity-based undersampling with the synthetic minority oversampling technique (SMOTE).
• Classification algorithm: Employs Rotation Forest incorporating support vector machine (SVM) subclassifications to segment feature spaces into multiple subsets for improved prediction.
• Cross-validation performance: Tenfold cross-validation yielded MCC values of 0.2287 (K), 0.3125 (P), 0.2787 (R), and 0.2814 (T) with False Positive rates of 0.2628, 0.1084, 0.1383, 0.1313 and False Negative rates of 0.2252, 0.0205, 0.0976, 0.0608 for K, P, R, and T respectively.
• Independent test performance: On an independent test dataset achieved MCC values of 0.6358 (K), 0.2910 (P), 0.4629 (R), and 0.3685 (T) with False Positive rates of 0.0165, 0.0203, 0.0188, 0.0094 and False Negative rates of 0.1026, 0.1875, 0.2037, 0.3333 for K, P, R, and T respectively.
• Functional characterization: Enhances computational functional characterization of proteins by providing residue-level predictions of carbonylation sites.

Scientific Applications:

• Aging: Supports identification of ROS-mediated carbonylation sites relevant to aging-related proteome changes.
• Neurodegenerative diseases: Enables study of carbonylation patterns implicated in neurodegenerative diseases.
• Inflammation: Facilitates analysis of protein carbonylation in inflammatory processes.
• Diabetes: Supports research into carbonylation-associated mechanisms in diabetes.
• Amyotrophic lateral sclerosis: Assists investigation of carbonylation roles in amyotrophic lateral sclerosis.
• Huntington's disease: Assists investigation of carbonylation roles in Huntington's disease.
• Tumor biology: Enables analysis of carbonylation in tumor-related biological studies.

Methodology:

Resampling using K-means similarity-based undersampling combined with SMOTE (SMOTE-KSU); classification via Rotation Forest with support vector machine (SVM) subclassifications; performance assessed by tenfold cross-validation and independent test set reporting MCC, false positive and false negative rates for K, P, R, and T sites.