The resulting scores are presented in Figure 2. The conclusion from the experiments is that although all estimation methods perform rather well, the centered least squares and centered ridge regression approaches seem to show the best performance.
Thus, for low s, both methods perform rather poorly and the CMH has a slightly higher AP in these cases.
However, with extremely common alleles, all methods perform rather poorly, with forward stepwise regression and the ATT doing slightly worse than the other methods.
The "make" method for strength testing was performed rather than the "break" method as it has been shown to have better reliability and provide more accurate measures [ 26, 27].
But this also applies to methods, which perform rather poor when a DC method is applied (e.g. WT-LDB and WT-BBC).
Interestingly, peptides 3 and 5 had previously performed rather poorly in direct thioester formation, and this new method presented a significant improvement.
Finally, despite these great efforts, existing methods perform badly for rather big patterns.
All methods performed well with no errors.
Both methods performed well on our data.
All methods performed similarly.
