Figure 1. HelixTree analysis spreadsheet

In Figure 1, we have a spreadsheet view of a HelixTree dataset. The BP (blood pressure) column has been made the dependent variable (denoted by the magenta color). We then select interactive tree analysis and begin our genetic and environmental subgroup analysis. HelixTree sweeps through all the independent variables to determine which one is most significant in terms of differences in average BP among subgroups, defined by the independent variables.

Figure 2. HelixTree recursive split tree with histogram

In Figure 2, a tree was created from a simulated data set of 1000 patients. In this tree, the dependent variable is blood pressure and the independent variables are factors such as age, sex and body-mass index (BMI) as well as several candidate genes. The first split in the tree shows that the main effect in blood pressure differs according to the age of the patient. For those patients 50 years of age and under, the most significant split is sex. The male (M) group has a mean blood pressure 4.9 points higher than that of females (F). From a genetics point of view, the picture becomes interesting when we find a subgroup of female smokers, who are homozygous 1_1 for Gene A, have a higher blood pressure than those with homozygous 2_2 or heterozygous 1_2. The histogram view inside Figure 2, shows the contribution of the two genetic subgroups relative to the parent population. Note how the mean response of the homozygous 1_1 Gene A patients (in purple) is shifted to the right.

Figure 3 shows the data points of a three way split on BMI. Blood pressure is on the Y-axis and the BMI is on the X-axis. By clicking on the vertical lines, the areas of the split can be changed. The slider bar at the lower right of the screen allows you to change the averages of the data points. Note there is an elevated blood pressure for patients whose BMI is between 22.8 and 25.2.

Figure 3. Three way split

Figure 4. Manually define split points

HelixTree allows you to interactively explore your dataset by building manual trees, choosing from a list of significant splitters, and discovering subgroups within your dataset, thereby predicting drug response, and finding subgroups within your dataset with good or bad responders to a particular treatment.