Genomic Anomaly Detection with Functional Data Analysis.

Area of Science:

• Genomics
• Evolutionary Biology
• Bioinformatics
• Artificial Intelligence

Background:

• Genetic variation is fundamental to understanding evolution.
• Machine learning and AI are increasingly used to study genomic footprints of evolution.
• Current methods often require prior knowledge of evolutionary drivers, limiting the discovery of unknown anomalous genomic regions.

Purpose of the Study:

• To introduce a novel, unsupervised approach for detecting anomalous genomic regions irrespective of their underlying causes.
• To provide a complementary method to existing simulation-based predictive modeling for evolutionary genomics.

Main Methods:

• Developed ANDES (ANomaly DEtection using Summary statistics), a suite of algorithms for unsupervised anomaly detection in genomic data.
• Employed statistical techniques to extract features, including derivatives of genetic variation across contiguous windows to capture linkage disequilibrium effects ('velocity' and 'acceleration').
• Trained models using these features to identify biologically significant or artifactual genomic regions.

Main Results:

• ANDES successfully identified anomalous genomic regions in human data, many of which corresponded to genes under selection (positive or balancing).
• Detected a non-uniform distribution of anomalies across the genome, enriched in specific chromosomal locations and sequence features (e.g., low GC content, repetitive sequences).
• Demonstrated the ability to flag potentially artifactual regions in addition to biologically significant ones.

Conclusions:

• ANDES provides a novel, model-agnostic framework for uncovering anomalous genomic regions.
• This approach is applicable to both model and non-model organisms, advancing the study of evolutionary genomics.
• The method enhances the discovery of novel evolutionary patterns and potential data artifacts.