In a polymorphism discovery project many candidate SNP are detected. Each of these SNP must be expertly evaluated and classified as true or false. ML was applied to expedite this manual step by filtering majority of the false predictions. The ML program C4.5 was applied to a set of carefully chosen features to build a SNP classifier from a training dataset along with expert decisions. The ML classifier has 97% overall accuracy (i.e., fraction of candidate SNP that were correctly classified) and 85 % positive predictive value (i.e., fraction of candidate polymorphisms being real).

ML has so far been only applied to polymorphism discovery from soybean amplified STS analyzed with PolyBayes. However, the optimized ML feature set and ML framework can be applied to other instances of polymorphism discovery.