yORIpred

yORIpred predicts species-specific yeast DNA replication origins (ORIs) using machine learning to identify initiation sites relevant to DNA replication and regulation of gene expression.

Key Features:

• Iterative feature representation: Constructs 40 optimal baseline models by exploring combinations of eight sequence-based encodings and five machine learning classifiers, including random forest, support vector machine (SVM), and extremely randomized trees.
• Ensemble-derived feature vectors: Concatenates predicted probabilities from the 40 baseline models to form novel feature vectors for downstream learning.
• Supervised iterative learning: Learns informative features across multiple sequential models in a supervised iterative mode to improve discrimination between ORIs and non-ORIs.
• Classifier comparison: Systematic analysis identified the SVM-learned feature representation as most effective at discriminating ORIs from non-ORIs, outperforming other classifiers evaluated.
• Benchmarking and validation: Performance was assessed through comprehensive benchmarking on the same training datasets and independent evaluations demonstrating superior and stable predictive accuracy relative to existing predictors.

Scientific Applications:

• Yeast ORI annotation: Identifies species-specific yeast DNA replication origins for genomic annotation and mapping of replication initiation sites.
• Study of replication and gene regulation: Enables investigation of how replication initiation sites influence DNA replication dynamics and regulation of gene expression.
• Experimental candidate selection: Provides predicted ORI candidates to support design and prioritization of experimental validation in genomic studies.

Methodology:

yORIpred builds 40 baseline models from combinations of eight sequence-based encodings and five machine learning classifiers (including random forest, SVM, and extremely randomized trees), concatenates the models' predicted probabilities into novel feature vectors for supervised iterative feature learning, performs systematic classifier comparison, and validates performance via benchmarking and independent evaluations.