Specific mutations in H5N1 mainly impact the magnitude and velocity of the host response in mice

Tchitchek, Nicolas; Eisfeld, Amie J; Tisoncik-Go, Jennifer; Josset, Laurence; Gralinski, Lisa E; Bécavin, Christophe; Tilton, Susan C; Webb-Robertson, Bobbie-Jo; Ferris, Martin T; Totura, Allison L; Li, Chengjun; Neumann, Gabriele; Metz, Thomas O; Smith, Richard D; Waters, Katrina M; Baric, Ralph; Kawaoka, Yoshihiro; Katze, Michael G

Missing expression values into each proteomics profile have been inferred using the 'impute.knn' k-nearest neighbor imputation method [66] available in the 'imputation' package [67] of the R suite [52] . In order to avoid inconsistent missing value imputations, the nearest neighbors have been restricted to proteomics profiles belonging to the same biological condition. The number of nearest neighbors used in the imputation (the k parameter) has been set to 2/3 of the number of samples of the biological condition. Out of the 1,069,500 (3,875x276) proteomic expression values, 450,814 (42.15%) had to be inferred. Biological conditions at the earliest time points post-infection showed the highest degree of imputation, while the biological conditions at the latest time points post-infection showed the lowest degree of imputation. When expression values for a protein having been found as missing in more than 2/3 of the number of samples of the biological condition, the expressions values have been set to 10E-4 in order to avoid infinite fold-changes values when comparing to other conditions. The whole compendium of proteomics expression profiles center based on the value of the 9 th decile and scale normalized [48] [49] [50] using the 'normalizeBetweenArrays' normalization method available in 'limma' package [51] of the R suite [52] .