Figure 3.

Raw data to alignment. (a) Raw data are processed using a level-finding algorithm (Supplementary Discussion) to identify transitions between levels in the current trace. A subsequent filter removes most repeated levels, which likely result from polymerase backsteps (indicated by `*'). (b) Extract the sequence of median current values of each level. (c) Align the current values to predicted values from the reference sequence using the quadromer map (Fig. 2a). Alignment is performed with a dynamic programming alignment algorithm similar to Needleman-Wunch alignment²⁰ (Supplementary Discussion). In some locations, levels are skipped in the nanopore read either owing to motions of the DNAP or errors made by the level finding algorithm. In other places, backsteps result in multiple reads of the same level. We determine read boundaries from the first and last matched levels in the reference sequence. Read boundaries are indicated by the blue lines. The above alignment had an estimated 6.4× 10⁻¹⁵ probability of false alignment.