Computationally Efficient Demographic History Inference from Allele Frequencies with Supervised Machine Learning.

Area of Science:

• Population genetics
• Computational biology
• Bioinformatics

Background:

• Inferring past demographic history from genomic data is crucial in evolutionary biology.
• The allele frequency spectrum (AFS) is a key summary statistic for demographic inference.
• Existing methods like dadi can be computationally intensive.

Purpose of the Study:

• To develop a more computationally efficient method for inferring demographic history.
• To maintain comparable accuracy to existing likelihood-based methods.
• To leverage machine learning for demographic inference.

Main Methods:

• Donni (demography optimization via neural network inference) utilizes Mean Variance Estimation neural networks.
• The method simulates AFS for various demographic models and parameters.
• Trained neural networks instantaneously infer model parameters from AFS.

Main Results:

• Donni achieves comparable accuracy to dadi for parameter and confidence interval estimates.
• The method significantly reduces the computational cost of demographic inference.
• Donni infers population size changes accurately and other parameters like migration rates fairly well.

Conclusions:

• Donni offers a computationally efficient alternative for demographic history inference.
• Supervised machine learning presents a promising approach for developing sustainable inference methods.
• This work facilitates large-scale genomic studies of population history.