Machine learning questions are hard to answer without knowing the details. What kind of data, processing steps, features, model you got would be important to know. Anyway, it seems you're assuming that only known regulatory regions should be responsible for determining the outcome. What about other regions ? Also if your assumption is true, then in effect what you're looking at is correlation between individual features and classification output (see my answer to your previous question). A predictive feature doesn't have to be correlated with output.