Popfinder: A Highly Effective Artificial Neural Network Package for Genetic Population Assignment

Area of Science:

• Genomics
• Bioinformatics
• Population Genetics

Background:

• Accurate genetic population assignment is crucial for ecology, conservation, and epidemiology.
• Low genetic differentiation poses challenges for traditional population assignment methods.
• Artificial neural networks show promise for genomic data-based population assignment.

Purpose of the Study:

• Introduce popfinder, a user-friendly Python pipeline utilizing artificial neural networks for genetic population assignment.
• Evaluate popfinder's performance on simulated and empirical data with varying genetic structures.
• Compare popfinder's accuracy and speed against existing population assignment software.

Main Methods:

• Developed a Python-based artificial neural network pipeline named popfinder.
• Tested popfinder with simulated genetic data under different gene flow scenarios.
• Applied popfinder to reduced-representation sequence data from three seabird species with weak population structure.

Main Results:

• Popfinder achieved high accuracy, precision, and recall in assigning individuals to source populations for most simulated and empirical datasets.
• Performance was robust even with low genetic differentiation, except in cases of extremely weak structure where comparators also struggled.
• Popfinder includes a perturbation ranking method for optimizing single nucleotide polymorphism (SNP) panels.

Conclusions:

• Popfinder is an effective and user-friendly tool for genetic population assignment, particularly valuable when genetic differentiation is low.
• The pipeline demonstrates high accuracy across diverse datasets, offering improvements over existing methods.
• Users should be mindful of potential issues like data leakage during model training to ensure reliable results.