De novo Nanopore read quality improvement using deep learning

BMC Bioinformatics

Table 1 Results from training and testing on different datasets

	LC train, LC test	LC train, HC test	HC train, LC test	HC train, HC test
Mean Sq. Error	0.00300	0.00447	0.00312	0.00391
Pearson	0.827	0.747	0.809	0.772
Spearman	0.805	0.795	0.778	0.802
Sensitivity	0.950	0.891	0.938	0.889
Specificity	0.681	0.734	0.681	0.751

“LC” is a low complexity, high coverage (140 × to 204 ×) community derived from 747,598 reads from only two species, Escherichia coli (204 × coverage) and Sphingomonas koreensis (140 × coverage). “HC” is a high complexity, low coverage (0.005 × to 64 ×) community derived from 260,930 reads from 26 species, described in [31]. The cutoff point for the sensitivity/specificity results was set at 0.8. We use the notation “LC train, HC test” to mean training the model on the LC data and testing it on the HC data

ISSN: 1471-2105