Fig 1.
Dengue proteins and antigenicity.
a) Structure of immature (left-half) and mature (right-half) dengue virion with viral RNA encapsulated. b) Organization of dengue proteins on a polyprotein tranlated from the viral RNA. c) Representation of two-dimensional antigenic map of dengue viruses. Viruses in each of the serotypes form antigenic clusters on the map. Antigenic distances can be measured from the map.
Fig 2.
Association between effect sites and known epitopes of neutralizing antibodies.
a) Number and percentage of sites with and without effects by whether or not they are part of known epitopes. Odds ratios were calculated by either considering epitopes of both human-derived monoclonal antibodies (hmAb) and murine-derived monoclonal antibodies (mmAb) and when only restricted to hmAb epitopes. b) Defining neighborhoods of known hmAb epitopes as positions within N sites away (linear distance), the probability of nonzero effect sites being within the neighborhood at random (red) are contrasted against the proportion of variable sites that were within the neighborhood (gray). c) Analogous analysis but with neighborhoods defined as being within X angstroms away from known hmAb epitopes (3-dimensional spatial distance). N = 0 and X = 0 were when the neighborhood was exactly at the reported epitope positions.
Fig 3.
Effects of substitutions in the envelope protein.
a) Substitutions with non-zero effect sizes with 95% interquartile range across the 100-fold Monte Carlo cross-validations as whiskers, median as points. Points are colored red if they match positions of known epitopes for monoclonal antibodies compiled in the DENVab database [24]. Gray vertical lines indicate positions with known human-derived monoclonal antibody (hmAb) epitopes, long if within site diversity exists in our dataset and short if not. b) Footprints of potently neutralizing hmAbs, colored red if the positions showed non-zero effects. c) Top and side views of the envelope protein structure with known epitopes colored red if estimated as non-zero effect, and gray if estimated as zero effect. Non-zero effect positions not matching reported hmAb epitopes are in black.
Fig 4.
Antigenic signal in each DENV protein.
a) Average within site variability in DENV proteins observed in the dataset. Bars were annotated with number of variable sites, total number of sites, and percentage of sites variable. b) Prediction performance of each DENV protein as observed (red) contrasted against expectations derived from random subsample of sites from any DENV protein of the same length (gray) and random down samples of sites from the envelope protein (E, blue). Points and lines are median and 95% interquartile range (IQR) of the root mean squared error (RMSE) evaluated under 100-fold Monte Carlo cross-validation. Length of the proteins are shown in parentheses. Only nonstructural protein 2A (NS2A) appeared to have better predictive performance than the expectations.
Fig 5.
Sites embedding antigenic signals beyond the envelope protein.
Prediction performance of downsampled NS2A sites concatenated with E when randomly downsampled to a) 60 sites and b) 30 sites contrasted against when concatenated with random sites from other proteins. c) Distribution of frequencies at which sites showed non-zero effect given being sampled in the two downsampling schemes. Black lines link frequencies of the same sites. d) Performance when concatenating the 62 sites which >99% of the times sampled was estimated to have non-zero effect size when adjusted for E in both schemes (red) compared against the same sites but permuted (yellow), and sites from other proteins of the same length (gray). Permutation was done by permuting residues observed at each site across viruses to conserve its diversity.