Prediction of antibody-antigen interaction based on backbone aware with invariant point attention

BMC Bioinformatics

Table 1 Performance on the AD-AbAg database

Method	\(\hbox {AUROC}^{a}\)	\(\hbox {AUPRC}^{a}\)	\(\hbox {Precision}^{a,b}\)	\(\hbox {Recall}^{a,b}\)	\(\hbox {Specificity}^{a,b}\)	F1 \(\hbox {score}^{a,b}\)
AbAgIPA-Paddle	0.721	0.781	0.810	0.555	0.865	0.654
AbAgIPA-Twins	0.701	0.770	0.7560	0.561	0.821	0.643
\(\hbox {SGPPI-Paddle}^{c}\)	0.683	0.705	0.653	0.564	0.700	0.603
\(\hbox {SGPPI-Twins}^{c}\)	0.659	0.681	0.737	0.423	0.790	0.473
AbAgInterPre	0.694	0.739	0.666	0.619	0.686	0.640

\(^{a}\)All metrics shown in the table are averages of 5-fold cross-validation;\(^{b}\)We use a decision boundary of 0.5 to determine TP(True Positive), FP(False Positive), TN(True Negative) and FN(False Negative); \(^{c}\)SGPPI-Twins are SGPPI’s primitive structure, the GCN network layer corresponding to the two inputs is parameter shared, and SGPPI-paddle is the GCN network layer corresponding to the two inputs side by side, and the parameters are unshared

ISSN: 1471-2105