MDD-carb

MDD-carb identifies protein carbonylation sites and characterizes motif signatures to predict carbonylated lysine (K), arginine (R), threonine (T), and proline (P) residues for studies of oxidative stress.

Key Features:

• Curated experimental dataset: Contains verified substrate sites from 241 carbonylated proteins comprising 332 K, 144 R, 135 T, and 140 P residue sites.
• Iterative statistical method: Employs an iterative statistical method to characterize motif signatures of carbonylation sites.
• Sequence-based features: Analyzes composition of twenty amino acids (AAC), composition of amino acid pairs (AAPC), position-specific scoring matrix (PSSM), and positional weighted matrix (PWM) around substrate sites.
• Dependency detection: Detects statistically significant dependencies in amino acid compositions around substrate sites to reveal potential motif signatures.
• Profile HMM training: Trains profile hidden Markov models (HMMs) on identified motif signatures and generates bit scores.
• SVM integration: Constructs an integrative support vector machine (SVM) model that combines HMM bit scores to balance predictive sensitivity and specificity.
• Performance evaluation: Reports balanced sensitivity and specificity demonstrated by cross-validation and independent testing.

Scientific Applications:

• Oxidative stress mapping: Mapping protein carbonylation sites as markers of oxidative stress in proteomic datasets.
• Motif discovery: Discovering motif signatures associated with carbonylation of K, R, T, and P residues.
• Disease research: Investigating protein carbonylation in metabolic and aging-related diseases including diabetes, chronic lung disease, Parkinson's disease, and Alzheimer's disease.
• Large-scale PTM analysis: Supporting large-scale proteomic analyses of irreversible oxidation by reactive oxygen species (ROS) as a post-translational modification (PTM).

Methodology:

The methodology uses an iterative statistical method to analyze AAC, AAPC, PSSM, and PWM around substrate sites, detect statistically significant dependencies, train profile HMMs to generate bit scores, and integrate those scores with a support vector machine (SVM), with performance assessed by cross-validation and independent testing.