Fig 7. Classification model for distinguish between early RA and not healthy but also not RA based on RPKMs.

'not RA' as present in the data, which is OA, arthralgia and undifferentiated arthritis, shortcut 'notRAarth'. The model consists of four rules, where three of them are shown here graphically. Panels A and B depict the first rule, where in B are only cases left which are higher than the threshold in A. The red shaded area in B shows the cases hit by the first rule, which are all earlyRA. Panel C shows the second and third rule, where the model output for the red shaded area is earlyRA and for the blue shaded area is notRAarth. The complete model as text is shown in D. The threshold values are rounded to one decimal place. The model corresponds to an accuracy of 86% at the 10-fold cross-validation (p-value 1.7*10⁻¹²). Variables for model-generation were pre-selected upon the intersection of single-variable comparisons. This pre-selection weakens the cross-validation as it is no part of it. This model is only intended for a distinction between early RA and other conditions as simple as possible based on gene expression. In total there were 95 samples as input data.