Table 5:

Comparison of normalized edit distance with different neural network architectures.

Architecture	Normalized edit distance
3 convolutional layers	0.4007 ± 0.0277
5 convolutional layers	0.3903 ± 0.0230
10 convolutional layers	0.3874 ± 0.0186
3 bidirectional recurrent layers	0.2987 ± 0.0221
5 bidirectional recurrent layers	0.2930 ± 0.0215
3 convolutional layers + 3 bidirectional recurrent layers	0.2011 ± 0.0252
5 convolutional layers + 5 bidirectional recurrent layers	0.2001 ± 0.0177

The normalized edit distance is the edit distance between predicted reads and labeled reads and normalized by segment length.