Predicting influenza antigenicity from Hemagglutintin sequence data based on a joint random forest method

Abstract

Timely identification of emerging antigenic variants is critical to influenza vaccine design. The accuracy of a sequence-based antigenic prediction method relies on the choice of amino acids substitution matrices. In this study, we first compared a comprehensive 95 substitution matrices reflecting various amino acids properties in predicting the antigenicity of influenza viruses by a random forest model. We then proposed a novel algorithm called joint random forest regression (JRFR) to jointly consider top substitution matrices. We applied JRFR to human H3N2 seasonal influenza data from 1968 to 2003. A 10-fold cross-validation shows that JRFR outperforms other popular methods in predicting antigenic variants. In addition, our results suggest that structure features are most relevant to influenza antigenicity. By restricting the analysis to data involving two adjacent antigenic clusters, we inferred a few key amino acids mutation driving the 11 historical antigenic drift events, pointing to experimentally validated mutations. Finally, we constructed an antigenic cartography of all H3N2 viruses with hemagglutinin (the glycoprotein on the surface of the influenza virus responsible for its binding to host cells) sequence available from NCBI flu database, and showed an overall correspondence and local inconsistency between genetic and antigenic evolution of H3N2 influenza viruses.

Machine learning prediction and experimental validation of antigenic drift in H3 influenza A viruses in swine

“…While there was considerable overlap between the positions in our model with the highest cumulative importance (Supplemental Table 3) compared to the positions in the JRFR algorithm (positions 62, 121, 131, 133, 135, 137, 142, 144, 145, 155, 156, 158, 159, 172, 173, 189, 193, 196, 276 ), the relative importance of these predictor features varied. Specifically, position 189 was the most important site in human H3 with ferret antisera, whereas our model identified position 145 as the most important position in swine H3 with swine sera (31). These differences of importance may be reflective of host specific interactions.…”

“…Our process included a robust analysis of prediction error and was able to identify the limits of the models. Using 10fold cross validation, our ensemble model had a higher RMSE when compared to a different machine learning approach developed for human IAV by Yao et al (2017) (31). This approach used a Joint Random Forest Regression (JRFR) algorithm that also included substitution matrices for predicting antigenic distances and had a RMSE < 1.0 (31).…”

“…This work adds to a growing body of literature that aims to quantitatively predict antigenic phenotypes of IAV from the sequence without requiring HI titers for each IAV strain (19,31,(42)(43)(44).…”

“…Specific amino acid substitutions were not weighted to minimize model assumptions, and prior research in human IAV has suggested that these approaches may add noise to analysis (30,31). All observed sitespecific amino acid substitutions in the reference data were identified and treated as bi-directional.…”

HMMPred: Accurate Prediction of DNA-Binding Proteins Based on HMM Profiles and XGBoost Feature Selection

“…Classification Algorithm. Two robust machine learning techniques, i.e., SVM and RF, are applied to perform the prediction of DBPs, which have been widely used for many classification tasks in the field of computational biology [43][44][45][46]. SVM is an outstanding classification method that is used to deal with a binary pattern recognition problem [47].…”