ASDEC

ASDEC detects selective sweeps across whole genomes by training convolutional neural networks on raw sequence data to identify regions under positive selection.

Key Features:

• Whole-Genome Scanning Capability: Performs comprehensive genome-wide scans to identify candidate genes and estimate the time and strength of selective sweeps.
• Direct Inference from Raw Sequence Data: Infers region characteristics directly from raw sequence data rather than relying on precomputed summary statistics, reducing sensitivity to confounding factors.
• Efficiency and Speed: Optimizes training and classification, achieving up to 10× faster training and 5× faster region classification compared to other CNN-based classifiers.
• Enhanced Sensitivity and Accuracy: Demonstrates up to 15.2× higher sensitivity, a 19.4-fold increase in success rates, and 4× greater detection accuracy relative to state-of-the-art methods.

Scientific Applications:

• Evolutionary biology: Enables detection of positive selection and study of adaptive evolution across genomes by identifying regions under selection.
• Parameter estimation: Provides estimates of the timing and strength of selection events to inform analyses of genetic adaptation.
• Population-scale genomic scans: Applies to population datasets such as human chromosome 1 from the 1000 Genomes Yoruba sample to detect known candidate genes.

Methodology:

Training a convolutional neural network (CNN) on raw genomic sequence data to recognize patterns indicative of selective sweeps while bypassing precomputed summary statistics.