3.1 The dataset
We analyzed the WCN (z-score) distribution of cellulase binding site residues that we can find from literature. Figure 2 shows the frequency of endo-glucanase binding residues (black) compared exo-glucanase binding residues (white). From the distribution, we can see WCN (z-score) of cellulase that most of the binding residues are between -1.6 ~ 0.9. After that, making statistics to measure sensitivity and specificity in order to decide the proper WCN threshold to predict binding site as follow.
3.2 The prediction performance
In this statistic, we calculate various WCN z- score threshold values to verify the sensitivity and specificity with respect to each threshold value. Besides, the threshold ranges from -1.6 to 0.9, increasing by 0.1 each time. If the residues we selected under threshold and also match the literature binding site residues, it is considered as “positive”; otherwise, it is considered as “negative”. Therefore each threshold value will produce a group of TPR and FPR, which decides a point on the diagram in Figure 3 and the list in Table 5. Figure 3(A)(B)(C) from top to the bottom shows all dataset cellulase diagram, endo-glucanase group diagram and exo-glucanase group diagram. Based on the measurement of sensitivity, specificity and the diagram
of a relationship TPR and FPR, we decide the suitable WCN z- score binding site threshold for endo-glucanases is < – 0.5 and for exo-glucanase is < -0.8.
3.3 Comprehend analysis of endo- and exo-glucanases
Despite the good prediction performance of WCN model, the sensitivity and specificity for predicting cellulase binding sites are need to improve, so we add a filter RSA to raise the values of specificity. The RSA threshold we selected (≥ 5%) based on Rost and Miller.20,21
3.3.1 Endo-glucanases
In this study, the endo-glucaase dataset we selected is as follows, PDB id 1TML, 2ENG, 1JS4 and 2NLR. The Figure 4(A) shows the WCN model of enzyme 1TML structure, Figure 4(B) shows the WCN z- score distribution of 1TML, Figure 4(C) compares Figure (D)(E) shows the experimental binding site residues colored in red, the residues under the WCN threshold (< -0.5) colored in orange and then selected residues that are exposed colored in orange, all of them are surface form and Figure (F) compares Figure (G)(H) shows the cartoon protein structure form, Figure 4(E)(H) means that we pick the residues that conform to WCN and RSA at the same time. And the comparison method WCN with WCN included RSA of sensitivity and specificity is shown in Table 6. Figure 5(A) to (H) shows the information of enzyme 2ENG structure like as Figure 4, the comparison method WCN with WCN included RSA of sensitivity and specificity is shown in Table 7. Figure 6(A) to (H) shows the information of enzyme 1JS4 structure like as Figure 4 enzyme 1TML, the comparison
method WCN with WCN included RSA of sensitivity and specificity is shown in Table 8. Figure 7 (A) to (H) shows the information of enzyme 2NLR structure like as Figure 4 enzyme 1TML, the comparison profile of endo-glucanase 2NLR method WCN with WCN included RSA of sensitivity and specificity is shown in Table 9.
Above-mentioned the relationship of performance, we combine two methods WCN and WCN include RSA, we can figure out the residues under WCN z-score threshold of enzymes we selected are much more than the method include RSA, although the sensitivity value will decrease, however we can lower the false positive value and enhance the true negative value then our specificity value will increase much more. It is clear that the binding site residues tend to have lower WCN z-score value and exposed according to our comparison with performance profile results.
3.3.2 Exo-glucanases
In this study next to the endo-glucanases, the exo-glucaase dataset we selected is as follows, PDB id 1CEL, 1QK2, 2HIS and 1EXP. Because of the enzyme 2HIS and 1EXP are in the same family, we select 1EXP for discussing only. The Figure 8(A) shows the WCN model of enzyme 1CEL structure, Figure 8(B) shows the WCN z- score distribution of 1CEL, Figure 8(C) compares Figure (D)(E) shows the experimental binding site residues colored in red, the residues under the WCN threshold (< -0.8) colored in orange and then selected residues that are exposed colored in orange, all of them are cartoon form, Figure 8(D)(E) means that we pick the residues that conform to WCN and RSA at the same time. And the comparison method WCN with WCN included RSA of sensitivity and specificity is shown in
Figure 8, the comparison method WCN with WCN included RSA of sensitivity and specificity is shown in Table 11. Figure 10(A) to (H) shows the information of enzyme 1EXP structure like as Figure 8 enzyme 1CEL, the comparison method WCN with WCN included RSA of sensitivity and specificity is shown in Table 12. Above-mentioned the relationship of performance, we also combine two methods WCN and WCN include RSA, we can figure out the residues under WCN zscore threshold (≤ -0.8) of enzymes we selected are much more than the method include RSA, although the sensitivity value will decrease, however we can lower the false positive value and enhance the true negative value then our specificity value will increase much more. It is clear that the binding site residues tend to have lower WCN z-score value and exposed according to our comparison with performance profile results.
Figure 11 shows the frequency of amino acid type in cellulase experimental binding site compared with our method of binding site prediction base on WCN including RSA, we expect that experimental and our work would be have similar amino acid type on binding substrates. However, the experimental frequency of hydrophobic, hydrophilic and charged amino acid type in endo-glucanases binding site are 35%, 35% and 31%; in exo-glucanases are 23%, 38% and 38%. And in our work, the frequency of hydrophobic, hydrophilic and charged amino acid type in endo-glucanases binding site are 41%, 34% and 25%; in exo-endo-glucanases are 45%, 24% and 31%. There is no significant correlation with the frequency of amino acid type in binding substrate between experimental and our work. However, we still can figure out the performance values of endo-glucanases and exo-glucanases show in Table 12, that values are increasing in specificity, through different WCN z- score threshold and RSA included. It means that enzymes are very specific, and the binding sites of enzyme are especially less flexible than other residues. The WCN is used for indicate
structure rigidity. However, there are complementary relationships between structural characteristics of binding sites and based on the method WCN. The WCN z- score threshold for endo-glucanases is larger than exo-glucanases, it means that most of endo-glucanases binding substrate structure are flexible, and exo-glucanases are more rigid for cellulose hydrolyze. It is reasonable that endo-glucanses need more space for hydrolyzing cellulose. Thus, using WCN and RSA may help understanding and finding binding site residues of cellulase as more as possible.