CL-Pore: Contrastive Learning for Accurate Nanopore Sequence Recognition

Nanopore sequencing represents a potent means for single-molecule biomolecular characterization, yet the precise identification of sequences from nanopore data remains a formidable challenge, attributed to the intrinsic noise and data complexity. This study introduces a cutting-edge contrastive learning-based methodology designed to significantly bolster the accuracy of nanopore sequence identification. By harnessing Oxford Nanopore Technologies (ONT) barcode data for initial pre-training, we designed a model referred to as CL-pore (Contrastive Learning for nanopore data). The efficacy of CL-pore is further enhanced through fine-tuning with downstream nanopore datasets. To augment the training process, we introduce a novel data augmentation strategy. This involves the inversion of signals, the strategic injection of noise and the cropping of signals, creating pairs of positive and negative samples that are instrumental for contrastive learning. Rigorous experimental assessments reveal that CL-pore markedly surpasses previous methods, setting a new benchmark for performance. It consistently outperforms existing state-of-the-art techniques in accuracy and robustness for nanopore sequence detection. The results underscore the power of contrastive learning to distill meaningful features from noisy nanopore data, thereby achieving unparalleled accuracy in sequence identification. This methodology not only advances the potential of nanopore sequencing technologies but also paves the way for their expanded use in genomic and molecular biology research, where accurate sequence detection is paramount.