Table 2.

Isoelectric point prediction accuracy on leave-out 25% datasets

Method	Protein dataseta				Method	Peptide datasetb
	RMSE	MAE	R 2	Outliersc		RMSE	MAE	R 2	Outliersc
IPC2.protein.svr.19	0.8479	0.5906	0.5934	247	IPC2.peptide.Conv2D	0.2216	0.1216	0.9761	2691
IPC2_protein	0.8608	0.6052	0.5748	251	IPC2.peptide.svr.19	0.2299	0.1155	0.9743	2490
IPC_protein	0.8677	0.6109	0.5760	250	IPC2_peptide	0.2482	0.1394	0.9700	3179
ProMoST	0.9113	0.6444	0.5183	263	Bjellqvist	0.4051	0.2836	0.9204	11639
Toseland	0.9278	0.6537	0.5095	250	Nozaki	0.4083	0.2673	0.9191	9837
Dawson	0.9365	0.6586	0.4977	263	DTASelect	0.4235	0.2796	0.9130	10606
Bjellqvist	0.9369	0.6536	0.5005	260	Thurlkill	0.4466	0.2535	0.9033	7182
Wikipedia	0.9484	0.6795	0.4860	262	Sillero	0.4747	0.2696	0.8907	7607
Rodwell	0.9579	0.6762	0.4706	262	Dawson	0.4910	0.2642	0.8831	6698
Grimsley	0.9588	0.6953	0.4779	265	Wikipedia	0.5178	0.2974	0.8700	8326
Lehninger	0.9617	0.6783	0.4607	266	Grimsley	0.5264	0.3796	0.8656	15956
Solomon	0.9631	0.6746	0.4606	272	Rodwell	0.5855	0.3429	0.8337	9857
pIR	1.0148	0.7556	0.4161	315	Toseland	0.5860	0.3896	0.8335	13152
Nozaki	1.0164	0.7219	0.3980	288	EMBOSS	0.5971	0.3557	0.8271	11022
Thurlkill	1.0250	0.7573	0.3948	302	PredpI-iTRAQ8	0.6302	0.3503	0.8027	12059
DTASelect	1.0278	0.7798	0.3947	319	PredpI-TMT6	0.6365	0.3518	0.7988	12135
EMBOSS	1.0498	0.7757	0.3734	308	PredpI-plain	0.6480	0.3710	0.7913	12813
Sillero	1.0519	0.7694	0.3461	308	IPC_peptide	0.7459	0.4860	0.7302	13599
Patrickios	2.3764	1.8414	<0	517	Solomon	0.7518	0.4929	0.7259	13777
PredpI-TMT6	NA	NA	NA	NA	Lehninger	0.7697	0.5209	0.7127	15200
PredpI-plain	NA	NA	NA	NA	pIR	0.8529	0.7303	0.6387	27158
PredpI-iTRAQ8	NA	NA	NA	NA	ProMoST	1.1026	0.7562	0.4104	18513
					Patrickios	2.0172	1.3927	<0	22818

^aProtein dataset consisting of 581 proteins (25% randomly chosen proteins, not used for the training or optimization).

^bPeptide dataset consisting of 29 774 peptides (25% randomly chosen peptides, not used for the training or optimization).

^cThe outliers were defined at 0.5 and 0.25 pH unit difference between the predicted and experimental pI thresholds for the protein and peptide datasets.

NA: The PredpI program was designed for peptides only within the 3.7–4.9 pH range; thus, for proteins, it returned 0 and could not be evaluated on the protein dataset.

New machine learning models developed in this study are in bold. First version of IPC (12) is underscored. Scores calculated after 10-fold cross-validation. Table is sorted by RMSD. For individual methods’ predictions, see Supplementary Data 2. For more details about the datasets, see Table 1.