GAEseq

GAEseq reconstructs genomic mixtures from high-throughput DNA sequencing data for single individual haplotyping and viral quasispecies reconstruction using a graph auto-encoder.

Key Features:

• Graph auto-encoder framework: Leverages structural properties inherent in sequencing data to model relationships between reads and variants.
• Neural network-based inference: Trains a neural network to mitigate sequencing errors and infer posterior probabilities of sequencing read origins.
• Read-origin assignment: Formulates and addresses the NP-hard problem of assigning reads to mixture components.
• Consensus-based reconstruction: Identifies and reconstructs mixture components by achieving consensus among reads inferred to originate from the same genomic component.
• Benchmark performance: Evaluated on realistic synthetic data and experimental datasets and reported to outperform state-of-the-art methods in haplotype assembly and viral community reconstruction.

Scientific Applications:

• Single individual haplotyping: Assembly of haplotypes from sequencing reads to resolve diploid or polyploid genomes.
• Viral quasispecies reconstruction: Reconstruction of viral community composition and strain sequences from mixed viral populations.
• Genomic mixture reconstruction: Deconvolution of mixed genomic samples into constituent components using read-origin inference.

Methodology:

Applies a graph auto-encoder neural network trained to infer posterior probabilities of read origins while mitigating sequencing errors and reconstructs mixture components by consensus among reads assigned to the same component.