Prediction of prokaryotic origins of replication (ORI)

Question

I want to predict origins of replication (ORI) on hundreds of prokaryotic genomes. The most straight-forward solution would be to use most commonly used tool, Ori-Finder.

It uses integrated gene prediction, analysis of base composition asymmetry, distribution of DnaA boxes, occurrence of genes frequently close to oriC regions and phylogenetic relationships to predict ORIs.

However, I came across a paper by Parikh et al which describes a single measure (correlated entropy measure / CEM) with which they achieved 100% ORI-identification on 500 genomes.

I like the CEM approach more than a close-source Windows-only black-box.

Parikh tested their algorithm on NCBI genomes. How are ORI annotated on these reference genomes? I couldn't find such an annotation. Does the sequence simply begin at the ORI in leading strand direction by convention? Were the ORI of the reference genomes identified by experiment?

Answer 1

I have not seen the papers but I know the technique SVM. This is a machine learning approach its pretty good, but has been around this the 1980s.

Its quite simple known ORIs are given with the associated genetic trait, e.g. GC skew, SVM will then regress the GC skew of a known ORI against the GC content of the genome, there will be a training, parameter optimization and test split in the data, typically 80:20:20. De novo data will then be feed into the trained algorithm, in the same way the test data was, and a prediction will be generated. There are weaknesses, the authors have used only one ML technique, typically numerous will be used e.g. SVM, random forest, lasso regression etc... In addition, the prediction accuracy for GC is low (75%), in addition 500 genomes isn't great for ML, you need 1000s.