A caveat of our modelling approach (using the Gaussian model) is that we could not include homopolymers of length zero due to the difficulty of modelling a normal which can only take on positive values. Based on the empirical error rates for over-called zero flows, we found this error rate (similar to longer homopolymers) increases by cycle, and in particular, is higher for PICS with a large positive coefficient in the model (i.e. PIC 12, followed by PICs 16, 23, 0 etc.), and very low for PICS with a negative coefficient (PIC 10, 13, 26 etc.) (Figure S9).
