MOCCA

MOCCA models cis-regulatory element (CRE) DNA sequences by applying motif occurrence combinatorics to analyze and predict promoters, enhancers, Boundary Elements (BEs), and Polycomb Response Elements (PREs).

Key Features:

• Motif occurrence combinatorics: Models distinct occurrence landscapes of sequence motifs across CREs.
• SVM-MOCCA: Hierarchical machine learning using Support Vector Machines to model local sequence composition at individual motif occurrences and combine occurrence-level models to predict regulatory elements.
• RF-MOCCA: Random Forest derivative implementing the hierarchical MOCCA approach.
• Support for motif formats: Accepts both IUPAC motifs and Position Weight Matrix (PWM) motifs.
• Negative data generation: Automatic generation of negative training data and a mode that operates with only positive examples, specified motifs, and a genome.
• Cross-validation evaluation: Performance assessed via cross-validation experiments on Drosophila PREs and BEs.
• Comparative performance: Demonstrates improved generalization over 4-spectrum and motif occurrence frequency Support Vector Machines and Random Forests, with RF-MOCCA achieving superior results in reported experiments.

Scientific Applications:

• PRE prediction in Drosophila: Prediction of Polycomb Response Elements demonstrated using SVM-MOCCA and RF-MOCCA.
• BE prediction: Prediction of Boundary Elements using the hierarchical MOCCA approach.
• Cis-regulatory element modeling: Modeling and distinguishing promoters, enhancers, BEs, and PREs based on motif occurrence patterns.

Methodology:

Hierarchical machine learning with SVM-MOCCA modeling local sequence composition at individual motif occurrences and combining occurrence-level models; RF-MOCCA as a Random Forest derivative; cross-validation for evaluation; support for IUPAC and PWM motifs; automatic negative training-data generation; and a mode operating with only positive examples, specified motifs, and a genome.