NanoDeep: a deep learning framework for nanopore adaptive sampling on microbial sequencing

Nanopore sequencers can enrich or deplete the targeted DNA molecules in a library through reversing the voltage across individual nanopores. However, it requires substantial computational resources to achieve lots of rapid operations in parallel at read-time sequencing. We aim to a deep learning framework, NanoDeep, to overcome these limitations through incorporating Convolutional Neural Network (CNN) and Squeeze and Excitation (SE). To this end, we generated a mock sample composited with Escherichia coli DH5a, Staphylococcus epidermidis, Roseomonas mucosa, Pseudomonas aeruginosa, Staphylococcus hominis, Neisseria gonorrhoeae, Staphylococcus aureus and HEK293T (human). Then, the samples were subjected to MinION sequencing. One run is with routine setting, half nanopore of another one run is with NanoDeep adaptive sequencing and the remaining one with routine setting. Those datasets were used to establish and evaluate the performance of NanoDeep model.