dnabinder

dnabinder predicts DNA-binding proteins from amino acid sequences using Support Vector Machine (SVM) models to support genome annotation and studies of gene regulation.

Key Features:

• Input: Accepts amino acid sequences as the basis for prediction.
• Machine learning model: Uses Support Vector Machine (SVM) classifiers to distinguish DNA-binding and non-DNA-binding proteins.
• Sequence composition features: Employs amino acid composition and dipeptide composition as feature sets.
• Evolutionary features: Incorporates Position-Specific Scoring Matrix (PSSM) profiles to include evolutionary information.
• Datasets — DNAset: Trained on DNAset containing 1153 DNA-binding proteins and 1153 non-DNA-binding proteins.
• Datasets — Extended set: Trained on an extended dataset comprising 146 DNA-binding chains/domains and 250 non-binding chains.
• Performance metrics: Reported accuracies include 72.42% (amino acid composition) and 71.59% (dipeptide composition) on DNAset, and 79.80% (amino acid composition) and 86.62% (with PSSM) on the extended set, with PSSM integration noted to raise accuracy from 72.42% to 74.22% in a reported comparison.
• Training data characteristics: Models trained on non-redundant protein sequences.
• Benchmarking: Evaluated on a blind dataset with reported superior performance relative to existing methods.

Scientific Applications:

• Genome annotation: Identification of DNA-binding proteins to assist functional annotation of genomes.
• Gene regulation research: Support for studies of transcription factors and other proteins involved in gene regulation.
• Domain-level prediction: Prediction of DNA-binding proteins and DNA-binding domains from sequence data.

Methodology:

SVM models were trained on non-redundant protein sequence datasets (DNAset and an extended set) using amino acid composition, dipeptide composition, and PSSM profiles as features and evaluated on a blind dataset.