Unsupervised Learning Approach for Comparing Multiple Transposon Insertion Sequencing Studies.

Area of Science:

• Microbiology
• Genomics
• Bioinformatics

Background:

• Transposon insertion sequencing (TIS) is crucial for bacterial genome-scale forward genetic screens.
• Comparing TIS data across different screens and organisms remains challenging.
• Existing methods lack robust approaches for meta-analysis of TIS datasets.

Purpose of the Study:

• To introduce a novel post-hoc analytic framework, comparative TIS (CompTIS), for meta-analysis of multiple TIS datasets.
• To enable comparison of TIS data across replicates and independent screens, including those from different organisms.
• To facilitate the identification of conserved and species-specific genes through comparative analysis.

Main Methods:

• Utilized unsupervised learning, specifically principal-component analysis (PCA) and clustering, for TIS data meta-analysis.
• Implemented screen-level PCA and clustering to assess variation and relatedness between TIS screens.
• Performed gene-level PCA to identify loci with concordant or discordant phenotypes across selected screens.

Main Results:

• CompTIS successfully identified variations and relatedness among different TIS screens and environments.
• Analysis of Vibrio intestinal colonization TIS data revealed pan-Vibrio genes and species-specific requirements.
• The framework demonstrated applicability to diverse TIS screens regardless of upstream analysis methods.

Conclusions:

• CompTIS provides a versatile and effective approach for comparing multiple transposon insertion sequencing screens.
• PCA-based analytics can be widely applied to compare diverse TIS screens, revealing conserved and divergent genetic requirements.
• The framework aids in identifying shared and distinct genetic factors influencing bacterial phenotypes across different conditions and organisms.