A universal computational model for predicting antigenic variants of influenza A virus based on conserved antigenic structures

Abstract

Rapid determination of the antigenicity of influenza A virus could help identify the antigenic variants in time. Currently, there is a lack of computational models for predicting antigenic variants of some common hemagglutinin (HA) subtypes of influenza A viruses. By means of sequence analysis, we demonstrate here that multiple HA subtypes of influenza A virus undergo similar mutation patterns of HA1 protein (the immunogenic part of HA). Further analysis on the antigenic variation of influenza A virus H1N1, H3N2 and H5N1 showed that the amino acid residues' contribution to antigenic variation highly differed in these subtypes, while the regional bands, defined based on their distance to the top of HA1, played conserved roles in antigenic variation of these subtypes. Moreover, the computational models for predicting antigenic variants based on regional bands performed much better in the testing HA subtype than those did based on amino acid residues. Therefore, a universal computational model, named PREDAV-FluA, was built based on the regional bands to predict the antigenic variants for all HA subtypes of influenza A viruses. The model achieved an accuracy of 0.77 when tested with avian influenza H9N2 viruses. It may help for rapid identification of antigenic variants in influenza surveillance.

Figure 1. Position-dependent entropy.

Moving average position information entropy (MAPIE) was calculated with a window size of 11 for HA1 protein of influenza A virus H1 (blue), H3 (red) and H5 (green). The amino acid positions are numbered according to influenza A virus H3.

Figure 2. The defined regional bands and their contribution to antigenic variation.

(A) Visualization of the ten regional bands in HA1 protein defined as described in the methods. Each regional band is differently colored in the 3-D model. (B) Pearson correlation coefficients (PCC) between the antigenic variation and the changes in each regional band, for subtype H1N1 (triangles), H3N2 (circles) and H5N1 (diamonds). (C) Correlations between the regional band’s contribution to the antigenic variation and their distances from the top of HA1. “SCC”, Spearman correlation coefficient.

Figure 3. The Receiver Operating Characteristic (ROC) curve of PREDAV-FluA in five-fold cross-validations.

The table under the curve summarizes the Area under Curve (AUC) data of the shown curve.