Dolbico

Dolbico analyzes DNA local conformations and provides the Database of Local Biomolecular Conformers (dolce) to store and classify dinucleotide torsion-angle conformers for characterization of sequence-dependent structural polymorphism.

Key Features:

• Comprehensive Structural Data: Houses information from crystallographic and NMR structures, analyzing over 1,480 DNA structures to capture a wide spectrum of local conformations.
• Automatic Classification Workflow: Applies machine learning combining k-nearest neighbors (k-NN) with non-hierarchical single-pass clustering algorithms to classify dinucleotide conformations or flag them as unclassifiable.
• Identification and Annotation of New Conformers: Identified and annotated six new DNA conformers, including examples from guanine quadruplexes and Holliday junctions.
• Fourier Averaging and Clustering: Employs Fourier averaging combined with clustering techniques to analyze dinucleotide torsion angles and delineate conformer classes.
• Machine Learning Approaches: Integrates supervised and unsupervised machine learning methods for classification and discovery from extensive structural datasets.
• Structural Alphabets and Interaction Matrices: Uses structural alphabets and interaction matrices to classify protein–DNA interactions and reveal interaction patterns.
• Sequence Preference Analysis: Identifies sequence preferences underlying sequence-dependent recognition and backbone flexibility.
• Water-mediated Contact Analysis: Reveals water-mediated contacts and distinct interaction patterns involving transcription factors, nucleases, and DNA grooves.

Scientific Applications:

• Enhanced Understanding of DNA Polymorphism: Characterizes the conformational space of the DNA backbone to inform studies of sequence-dependent structural polymorphism.
• Sequence Preferences and Recognition Patterns: Provides insights into sequence-dependent recognition by identifying sequence preferences associated with specific conformers.
• Protein–DNA Interaction Analysis: Supports investigation of protein–DNA recognition principles by classifying interactions with structural alphabets and interaction matrices, highlighting differences among transcription factors, nucleases, and DNA grooves.

Methodology:

Uses Fourier averaging and clustering of dinucleotide torsion angles together with machine learning methods — including k-nearest neighbors (k-NN), non-hierarchical single-pass clustering, and supervised and unsupervised learning — to classify and identify DNA conformers.