polyapred

polyapred predicts polyadenylation signals (PAS) in human DNA sequences using Support Vector Machine (SVM) models to identify mRNA polyadenylation sites that affect mRNA regulation and stability.

Key Features:

• SVM-based prediction: Uses Support Vector Machine (SVM) models to classify PAS locations in human DNA sequences.
• Sequence context: Analyzes 100 nucleotides upstream and downstream of candidate PAS sites.
• Split nucleotide frequency technique: Implements a split nucleotide frequency technique to enhance prediction accuracy.
• Nucleotide frequency features: Evaluates mononucleotide, dinucleotide, trinucleotide, and tetranucleotide frequencies.
• Feature combinations: The best-performing model combines dinucleotide, second-order dinucleotide, and tetranucleotide frequencies.
• Performance metrics: Reports Matthews correlation coefficients (MCC) of 0.58 (mononucleotides), 0.69 (dinucleotides), 0.70 (trinucleotides), 0.69 (tetranucleotides), and 0.72 for the combined model.
• Precision and sensitivity: Achieves precision rates ranging from 75.8% to 95.7% and a sensitivity of 57% on independent datasets.
• Comparative performance: Demonstrates superior performance compared to existing methods on independent datasets.

Scientific Applications:

• PAS identification: Identification of polyadenylation signals in human genomic sequences.
• mRNA regulation studies: Investigation of mRNA regulation and stability linked to polyadenylation sites.
• Disease-associated mutation analysis: Analysis of mutations in PAS implicated in various diseases.
• Method benchmarking: Evaluation and comparison of PAS prediction methods on independent datasets.

Methodology:

Analyzes 100 nucleotides upstream and downstream of candidate PAS sites, computes split nucleotide frequencies (mononucleotide, dinucleotide, trinucleotide, tetranucleotide), and applies Support Vector Machine (SVM) models; the best model combines dinucleotide, second-order dinucleotide, and tetranucleotide frequencies.