NEURON-IDS: One-Shot Neural Error Correction for DNA Storage Sequences

DNA-based data storage offers unprecedented information density and durability, but its practical implementation is hindered by sequencing errors that can corrupt stored data. Traditional error correction methods using minimum edit distance decoding struggle with the complexity of multiple concurrent errors. We present NEURON-IDS (NEUral Recurrent Optimizer Network for Insertion, Deletion, and Substitution error correction), a novel neural framework that combines specialized bidirectional recurrent neural networks to handle insertions, deletions, and substitutions in DNA sequences. Our architecture employs parallel error-specific submodules with an innovative context fusion mechanism to effectively correct multiple simultaneous errors. Experimental results demonstrate that NEURON-IDS achieves up to a 80.3% reduction in nucleotide error rates compared to conventional minimum edit distance decoding approaches when handling sequences with multiple errors, particularly excelling in scenarios with four concurrent errors. By learning complex sequence patterns beyond simple edit distance metrics, NEURON-IDS represents a significant advancement toward reliable DNA-based data storage systems.