GoFAE-DND

GoFAE-DND detects non-canonical (non-B) DNA structures from Oxford Nanopore sequencing by framing detection as a novelty-detection problem and using an autoencoder regularized with goodness-of-fit tests to identify deviations in nanopore translocation time signals.

Key Features:

• Novelty-detection framing: Frames non-B DNA identification as a novelty-detection problem to distinguish anomalous conformations from canonical B-DNA.
• GoFAE autoencoder: Implements an autoencoder model named GoFAE-DND (Goodness-of-Fit AutoEncoder for Non-Canonical DNA Detection) that integrates GoF tests as a regularizer.
• Discriminative loss: Uses a discriminative loss function that encourages poor reconstruction of non-B DNA to facilitate anomaly separation.
• Gaussian GoF tests and P-values: Optimizes Gaussian goodness-of-fit (GoF) tests within the model to compute P-values indicating deviations from expected distributions.
• Nanopore translocation-time signal: Leverages differences in Oxford Nanopore sequencing translocation timing between non-B and B-DNA bases as the primary detection signal.
• Whole-genome validation: Validated on whole-genome Oxford Nanopore sequencing of the NA12878 reference sample, revealing significant translocation timing differences between non-B and B-DNA.
• Comparative benchmarking: Performs comparative analyses against other novelty detection methods using experimental datasets.
• Synthetic-data simulator: Uses a translocation time simulator to generate synthetic translocation-time data for benchmarking and validation.

Scientific Applications:

• Non-B DNA detection: Detection and mapping of non-canonical (non-B) DNA structures in genomic sequences using nanopore translocation-time signals.
• Conformational characterization: Characterization of differences in nanopore translocation timing between non-B and canonical B-DNA bases.
• Method benchmarking: Benchmarking and validation of novelty-detection algorithms on experimental and simulated nanopore translocation-time data.

Methodology:

Frames detection as a novelty-detection problem and trains an autoencoder model (GoFAE-DND) that integrates goodness-of-fit (GoF) tests as a regularizer, employs a discriminative loss to promote poor reconstruction of non-B DNA, optimizes Gaussian GoF tests to produce P-values, and evaluates performance on Oxford Nanopore translocation-time signals using experimental NA12878 data and synthetic data from a translocation time simulator.