Digital Image Processing to Detect Adaptive Evolution.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Machine learning and image processing have revolutionized the detection of genomic regions under natural selection.
• Traditional methods relied on population-genetic summary statistics, which are limited by specific genomic pattern expectations.
• Recent advances allow for automatic feature extraction from image representations of genomic data using convolutional neural networks.

Purpose of the Study:

• To evaluate the efficacy of alpha-molecule techniques for feature extraction from image representations of haplotype alignments.
• To assess the performance of these techniques in detecting signatures of hard and soft selective sweeps.
• To compare the interpretability and performance of alpha-molecule-based models with contemporary deep learning approaches.

Main Methods:

• Utilized digital image processing methods, specifically alpha-molecules (wavelet and curvelet decomposition), to extract features from image representations of haplotype alignments.
• Applied linear and nonlinear machine learning classifiers to the extracted features.
• Generated simulated genomic data for training and testing machine learning models.

Main Results:

• Alpha-molecule techniques achieved high true positive rates and accuracy in detecting hard and soft selective sweep signatures.
• The developed models demonstrated ease of visualization and interpretation.
• Performance rivaled that of current deep learning approaches for sweep detection.

Conclusions:

• Alpha-molecule techniques provide a powerful and interpretable approach for detecting selective sweeps in genomic data.
• These methods offer a competitive alternative to deep learning for analyzing genomic images.
• Further application of these image processing techniques can advance the study of natural selection in genomics.