Predicting Cross-Species Infection of Swine Influenza Virus with Representation Learning of Amino Acid Features

Figure 1

Flowchart of representation learning of amino acid features. After data cleaning, 36 signature amino acid positions based on entropy were screened. Six encoding algorithms with the change of parameter were used to explore the key information. All the 64 descriptors in the feature pool were used to train and predict with the RF models, and two types of predictions were achieved to be further optimized. Each swine virus was eventually represented by two optimized feature vectors, “class” and “prob.” Finally, the predictive models were built and compared.