Figure 3
Identification of those genetic variants that promised to provide the comparatively best distinction between the two patient groups. The bar plot shows the absolute group size standardized differences of the occurrences of each variant between the two patients groups. The 34 bars left of the solid vertical line indicate the genetic variants found by a computed ABC analysis (right upper part of the figure) to provide a statistically valid set of markers to be included in a complex biomarker. At the upper right corner is the ABC plot14 of the cumulative distribution function of the additional fraction of the absolute group differences in allelic numbers (solid curved line). In addition, ABC plots of the identity distribution, xi=constant (dotted line, that is, each genetic variant occurs at the same number in each group), and of the uniform distribution in the data range (dotted line, that is, each variant allele had the same chance to occur) are shown. Compared with the latter distributions, the solid curved line clearly indicates a highly unequal distribution of the group differences in the numbers of variant alleles. Further marks in this plot consist of a light grey and a darker grey star denoting the so-called Pareto and BreakEven points, respectively. The Pareto point A (Ax, Ay) is the point at the smallest distance (left oblique black line) to the ideal point at xy where the effort would be zero to obtain the whole yield. The BreakEven point B (Bx, By) marks the point on the ABC curve where its slope, dY/dE, equals 1, that is, the so-called profit gain dABC equals 1. Beyond that point, more information can only be gained with inadequately high efforts. The ABC analysis comes from economical informatics and, in the present context, aims at identifying the most informative genetic variant for group classification by dividing the 152 variants into 3 distinct subsets. Set A should contain the ‘critical few’, that is, those elements that allow obtaining a maximum of yield with a minimal effort.14 Set B comprises those elements where an increase in effort is proportional to the increase in yield. In contrast, set C contains the ‘trivial many’, that is, those elements that are not worth to be considered a biomarker. As a result, set A was used to establish a subsymbolic classifying biomarker.
