McPromoter

McPromoter predicts eukaryotic transcription start sites by integrating interpolated Markov chain sequence models with Gaussian-distributed DNA physical properties (bendability and GC content) to improve promoter identification.

Key Features:

• Probabilistic Modeling: Employs interpolated Markov chains to model sequence likelihoods capturing complex dependencies within DNA sequences.
• Integration of Physical Properties: Incorporates DNA bendability and GC content modeled with Gaussian distributions into the predictive framework.
• Joint Sequence/Profile Models: Represents promoters as sequences of consecutive segments characterized by joint likelihoods that combine sequence data and profiles of physical properties.
• Background Modeling: Utilizes two distinct joint sequence/profile models to differentiate coding and non-coding sequences, each comprising a mixture of sense and anti-sense submodels.

Scientific Applications:

• Transcription start site identification: Improves detection of eukaryotic transcription start sites by combining sequence and physical-property information.
• Promoter annotation in eukaryotic genomes: Aids annotation of promoter regions in genomic studies by modeling sequence and structural profiles.
• Gene regulation and expression studies: Supports analyses of regulatory elements and gene-expression mechanisms through more accurate promoter localization.
• Benchmarking and model comparison: Demonstrated reduction of false positives by about 30% on a large Drosophila test set compared to models based solely on sequence likelihoods.

Methodology:

Constructs probabilistic models using interpolated Markov chains for sequence likelihoods and Gaussian distributions for DNA physical properties; models promoters as consecutive segments with joint sequence/profile likelihoods and models background with two joint sequence/profile models (coding vs non-coding), each a mixture of sense and anti-sense submodels.