Fig. 4: Case study. | Nature Communications

Fig. 4: Case study.

From: Multi-modal deep learning enables efficient and accurate annotation of enzymatic active sites

Fig. 4: Case study.

a Visualization of the enzyme active site annotation results of the EasIFA model in the test set. The left panel shows EasIFA’s annotation results for the active site of protein tyrosine phosphatases (UniProt ID: Q4G0W2, EC Number: 3.1.3.48), with the table below detailing the predicted active site amino acids information. In the ‘Correct’ column, ‘Y’ indicates that the prediction completely coincides with the recorded results in the dataset, while ‘R’ represents redundant results not recorded in the dataset (same below). The right panel displays EasIFA’s annotation results for the active site of carnosine N-methyltransferase (UniProt ID: P53934, EC Number: 2.1.1.22), with the table above showing the predicted active site amino acids information. b Visualization of the enzyme active site annotation results of the EasIFA model in the test set, which contain TIM Barrel structures but have completely different functions. The left panel shows the annotation results of the active site for Malate synthase A (UniProt ID: P08997, EC Number: 2.3.3.9), and the table below provides detailed information of the predicted active site amino acids. The right panel shows the annotation results of the active site for Ketose 3-epimerase (UniProt ID: A0A1L7NQ96, EC Number: 5.1.3.-), and the table above shows the information of the predicted active site amino acids. c Confusion matrix obtained using EasIFA on a test set of pseudoenzyme-enzyme pairs. d Visualization of a differentiated pair of kinase and pseudoenzyme using EasIFA. For the kinase (upper), serine/threonine-protein kinase BIK1 (UniProt ID: O48814, EC Number: 2.7.11.1), EasIFA accurately predicted all the active sites recorded in UniProt. For the pseudoenzyme (lower), inactive serine/threonine-protein kinase BKN1 (UniProt ID: Q9LFL7), EasIFA did not predict any catalytic active sites. The E-value between the two proteins is \(1.64\times {10}^{-58}\), calculated by BLASTp.

Back to article page