Prediction Errors vs. Upper-bound Prediction Error. The prediction errors of the three programs, Orphelia, MGA and GeneMark, and their upper-bound, for fragment lengths 100 to 700 bp. For 200-400 bp, Orphelia has the highest prediction errors, while MGA has the highest for longer read lengths. GeneMark has the worst prediction error for 100 bp read lengths. Also, the upper-bound analysis (which marks a read correct if at least one of the algorithms correctly predicts it) significantly reduces the error rate, showing promise for combining the methods.