Combination therapies can overcome antimicrobial resistance (AMR) and repurpose existing drugs. However, the large combinatorial space to explore presents a daunting challenge. In response, machine learning (ML) algorithms are being applied to identify novel synergistic drug interactions from millions of potential combinations. Here, we compare ML-based approaches for combination therapy design based on the type of input information used, specifically: drug properties, microbial response and infection microenvironment. We also provide a compilation of publicly available drug interaction datasets relevant to AMR. Finally, we discuss limitations of current ML-based methods and propose new strategies for designing efficacious combination therapies. These include consideration of in vivo conditions, design of sequential combinations, enhancement of model interpretability and application of deep learning algorithms.