Staged heterogeneity learning to identify conformational B-cell epitopes from antigen sequences

BMC Genomics

Table 4 Prediction performance of single propensities by LOOCV using Random Forest (column 2–5) and SVM (column 6–9)

Rank	Propensity	Recall	Precision	F-score	Propensity	Recall	Precision	F-score
Best1	P209	0.618	0.188	0.289	P206	0.720	0.177	0.284
Best2	P206	0.646	0.174	0.274	P209	0.630	0.179	0.279
Best3	P170	0.560	0.173	0.265	P208	0.724	0.167	0.271
Worst3	P122	0.362	0.165	0.227	P84	0.061	0.215	0.095
Worst2	P46	0.341	0.170	0.227	P166	0.090	0.100	0.095
Worst1	P44	0.283	0.169	0.212	P142	0.084	0.090	0.087
Average		0.537	0.165	0.252		0.407	0.163	0.219

The propensity 208 (secondary structure) ranks fourth in RF, and has a recall of 0.670, precision of 0.164, and F-score of 0.264. Propensity 207 (conservation score) ranks 201 and 100 in RF and SVM, and has an F-score of 0.239 and 0.235 respectively. Explanations to other important propensities: 209-PSSM, 206-ASA, 170-Optimized beta-structure-coil equilibrium constant (AAIndex ID: OOBM850101). AAIndex ID for worst propensities: 122-CHAM830107, 46-FAUJ880111, 44-FAUJ880112, 84-AURR980110, 166-OOBM770104, 142-GEOR030104. Antigens without a certain propensity were ignored in this table

ISSN: 1471-2164