Timesweeper

Timesweeper identifies selective sweeps in genomic time-series data using convolutional neural networks to detect polymorphisms under positive selection and estimate selection coefficients.

Key Features:

• Convolutional Neural Network Architecture: Timesweeper employs a one-dimensional convolutional neural network to analyze genomic time-series and detect polymorphisms under selection.
• Simulation-Based Training: Training datasets are generated by simulations under demographic models relevant to the study to calibrate the neural network.
• Detection of Targeted Polymorphisms: The method identifies polymorphisms directly targeted by selective sweeps, including ongoing and completed sweeps.
• Selection Coefficient Estimation and Robustness: Timesweeper estimates selection coefficients and maintains performance across varied simulated demographic scenarios and sampling conditions.
• Time-Series Sampling Support: The approach is designed to handle multiple genomic samplings from a population across different time points.

Scientific Applications:

• Experimental evolution and serial sampling: Detecting selection in experimentally evolved populations sampled repeatedly over time.
• Ancient DNA time-series: Inferring recent positive selection from temporal samples of ancient DNA.
• Rapidly reproducing or repeatedly sampled populations: Studying selection in organisms with short generation times or populations where serial sampling is feasible.

Methodology:

A one-dimensional convolutional neural network is trained on simulated genomic time-series datasets generated under specified demographic models to classify sites and estimate selection coefficients.