Pandora: a tool to estimate dimensionality reduction stability of genotype data.

Area of Science:

• Population Genetics
• Bioinformatics
• Computational Biology

Background:

• Genotype datasets are high-dimensional, posing challenges for population structure analysis.
• Dimensionality reduction techniques are crucial for inferring population membership but require robust uncertainty estimation.
• Current methods lack a standardized approach for assessing the stability of genotype data embeddings.

Purpose of the Study:

• To develop a stability estimation framework for genotype data.
• To quantify the uncertainty inherent in dimensionality reduction of genetic data.
• To provide a tool for validating population structure inferences.

Main Methods:

• Developed Pandora, a bootstrapping-based stability estimation framework.
• Implemented overall embedding stability scores and per-individual support values.
• Integrated a k-means clustering approach for cultural group assignment uncertainty.

Main Results:

• Pandora effectively quantifies embedding stability and individual support.
• The framework demonstrates utility on both simulated and empirical genotype datasets.
• Pandora provides a reliable method for assessing uncertainty in population genetic analyses.

Conclusions:

• Pandora offers a robust solution for uncertainty assessment in genotype data dimensionality reduction.
• The framework enhances the reliability and interpretability of population structure studies.
• Pandora is publicly available, promoting reproducible research in population genetics.