Shown in plot A of the above figure are two distributions of margins of SVM models for an individual variable (say Gene A): the black peak (denoted by 1) is the distribution of a population of models which contain Gene A, while the white peak (denoted by 0) shows the distribution of a different population of models which DO NOT include Gene A. As is known, the larger the margin of a SVM model is, the better prediction accuracy it would have. Therefore, based on Plot A, it can be concluded that including Gene A in a model would ON AVERAGE increase the margin of a SVM model and hence Gene A is considered to be informative.
In contrast, a gene with overlapping margin distribution as shown in Plot B, would be considered to be uninformative since including this gene in a model (Peak 1) decrease the margin comapred to the SVM models without including this gene (Peak 0).
If you use this method, please cite the following paper:
H.-D. Li, Y.-Z. Liang, Q.-S. Xu, et al., Recipe for Uncovering Predictive Genes using Support Vector Machines based on Model Population Analysis, IEEE/ACM T Comput Bi, 8 (2011) 1633. PDF