Translational adaptation of human viruses to the tissues they infect

Xavier Hernandez-Alias; Martin Schaefer; Luis Serrano

To evaluate the adaptation of the viral proteins to the SDAw of human tissues, we computed their average SDA to each of the 23 TCGA tissues (Sup. Table 3 ). Using the set of 182 tropism-defined viruses, we had a total of 2891 viral proteins. Taking the 23 tissue-specific SDAs as features, we applied a Random Forest (RF) Classifier, populated with 100 decision trees, using the scikit-learn package 69 . Therefore, for each of the six viral tropisms, we developed a model for predicting the tropism-positive versus tropism-negative proteins based on the translational adaptation across tissues. Given that the size of the tropism-positive and tropism-negative groups were often unbalanced, we iteratively sampled equal-sized groups, for n=100 iterations. Furthermore, we validated the results with a stratified 5-fold crossvalidation.