This document discusses using machine learning models to predict pathogen phenotypes like drug resistance and virulence from genomic data. It presents results from models trained to predict HIV drug resistance from protease sequences with good performance. The models captured the temporal emergence of resistance after new drug approvals. The approach aims to establish genomic surveillance pipelines for pathogens by computing a "risk factor" from sequence data to help monitor zoonotic pathogens. Challenges include mapping genotypes to phenotypes given epistasis and lack of data. The vision is to assay pathogen populations to build genotype-phenotype datasets to train models that can quantify risk profiles and inform interventions.