Analysis of the Impact of Potential Bias in the SwissProt dataset.

Bias was assessed in terms of impact on the fit statistics for power-law behaviour in the tail of the unique amino acid count distribution, which is addressed in detail in the following sections as prediction P1, (see also Fig 1A). Datasets Psub (ai = 20 to 30), Pexp (ai = 20 to 31) and Pall (ai = 20 to 37), were analyzed as extracted from SwissProt and are described in Methods. The fit statistics for the power-law tail were then compared with the equivalent fit statistics on datasets corrected for potential bias in treatment of the initiating methionine (M-fixed), inclusion or exclusion of signal peptides (No peptides) and Monte-Carlo exploration of the ambiguity of unique amino acid counts as described in Methods. The fit statistics are remarkably resilient with respect to all three different possible sources of bias, and the robust linearity of the power-law tail in all conditions is emphasized by the high values of the adjusted R2.