A very good separation between classes can sometimes be provided by looking at a single feature variable (gene) so that the classification error rate is difficult to reduce further by including other (probably quite significant) variables in the rule.
1
—
—