Evaluation of genomic island predictors using a comparative genomics approach

Area of Science:

• Genomics
• Bioinformatics
• Microbial genetics

Background:

• Genomic islands (GIs) are crucial for microbial adaptations, often originating from horizontal gene transfer.
• Automated detection methods for GIs typically rely on sequence composition biases.
• Robust evaluation requires GI datasets independent of sequence composition analysis.

Purpose of the Study:

• To develop a novel comparative genomics approach for identifying genomic islands.
• To construct independent datasets for evaluating existing genomic island prediction tools.
• To assess the accuracy of sequence composition-based genomic island predictors.

Main Methods:

• Developed IslandPick, a comparative genomics approach for identifying genomic islands.
• Implemented automated selection of relevant comparative genomes.
• Created positive (unique regions) and negative (conserved regions) datasets for evaluation.

Main Results:

• IslandPick demonstrated high accuracy in identifying genomic islands compared to curated lists.
• Evaluated multiple sequence composition-based prediction tools using the constructed datasets.
• SIGI-HMM showed the highest precision, while AlienHunter had the highest recall but lowest precision.

Conclusions:

• The comparative genomics approach (IslandPick) provides accurate datasets for evaluating genomic island predictors.
• SIGI-HMM and IslandPath/DIMOB offer the highest overall accuracy among tested tools.
• This study presents the first independent evaluation of sequence composition-based genomic island predictors using diverse, non-artificial datasets.