Sigma70Pred

Sigma70Pred predicts sigma70 promoters in prokaryotic genomes using an SVM-based model trained on sequence-derived correlation and composition features to support studies of transcriptional regulation.

Key Features:

• Dinucleotide Auto-Correlation: Sequence-derived descriptor capturing correlations between dinucleotide positions.
• Cross-Correlation: Descriptor capturing cross-correlations between nucleotide properties across the sequence.
• Auto Cross-Correlation: Descriptor combining autocorrelation and cross-correlation information from sequences.
• Moran Auto-Correlation: Spatial autocorrelation descriptor used to characterize nucleotide property distributions.
• Normalized Moreau-Broto Auto-Correlation: Normalized auto-correlation descriptor applied to sequence properties.
• Parallel Correlation Pseudo Tri-Nucleotide Composition: Pseudo tri-nucleotide composition descriptor capturing higher-order nucleotide correlations.
• SVM-based model: Support Vector Machine classifier trained on selected sequence features for sigma70 promoter prediction.
• Feature selection: Model optimization performed using a subset of 200 selected features.
• Training performance: Achieved accuracy of 97.38% and AUROC of 0.99 on the training dataset.
• External validation: Tested on RegulonDB10.8 independent dataset (1,134 sigma70 promoters, 638 non-sigma70 promoters) with accuracy 90.41% and AUROC 0.95.

Scientific Applications:

• Sigma70 promoter identification: Predicts sigma70 promoters in prokaryotic genomes to aid annotation of promoter elements.
• Transcriptional regulation analysis: Supports studies of transcriptional regulation of housekeeping genes in prokaryotes.
• Benchmarking and validation: Enables benchmarking of promoter-prediction methods using RegulonDB10.8 as an external reference dataset.

Methodology:

Sequence features were computed using Dinucleotide Auto-Correlation, Cross-Correlation, Auto Cross-Correlation, Moran Auto-Correlation, Normalized Moreau-Broto Auto-Correlation and Parallel Correlation Pseudo Tri-Nucleotide Composition; a subset of 200 features was selected and used to train an SVM-based classifier, which was validated on an independent RegulonDB10.8 dataset (1,134 sigma70 promoters, 638 non-sigma70 promoters).