Parameterization of disorder predictors for large-scale applications requiring high specificity by using an extended benchmark dataset

Table 7 Performance benchmark with the SL dataset under parameters that produced the highest Matthews Correlation Coefficient (MCC).

Method	threshold	sensitivity	specificity	MCC	PE
DISOPRED2	0.05	0.645	0.897	0.567	0.541
IUPred long	0.48	0.627	0.907	0.564	0.534
IUPred short	0.41	0.649	0.877	0.546	0.526
CAST	24	0.578	0.908	0.522	0.485
SEG45	3.45;3.75	0.582	0.841	0.442	0.423
DisEMBL Rem465	1	0.348	0.969	0.418	0.317
SEG25	3.05;3.35	0.460	0.885	0.387	0.345
DisEMBL Coils	1.8	0.515	0.835	0.373	0.350
SEG12	2.35;2.65	0.282	0.943	0.308	0.225
DisEMBL Hotloops	2.3	0.306	0.928	0.304	0.233

Predictors were run under parameters that produced the highest MCC over the SL dataset to benchmark their maximally possible performance. Rankings by MCC and PE differ only slightly. Interestingly, the identified optimal parameters (in regard to MCC performance over our dataset) often differed from the default parameters of the respective programs, except for DISOPRED2.

ISSN: 1471-2164