Antibiotic resistance poses a significant and growing threat to public health, with an estimated annual toll of 10 million deaths by 2050 without intervention [1]. At the same time, traditional antibiotics discovery pipelines have become less productive [2,3]. Consequently, there is a need for new methodologies that can discover new antibiotics quickly and at a low cost, and the application of computational approaches and machine learning (ML) is a promising avenue. Metallo-β-lactamases (MBLs) are a key target for combating antibiotic resistance because they provide resistance to last resort antibiotics such as carbapenems. In this contribution, we introduce an ML-based pipeline for the optimisation of MBL inhibitors using a combination of physics-informed, traditional ML methods and generative models. Our results demonstrate that our pipeline can optimise hit compounds for multiple objectives, such as target inhibition and microbiological activity. In future work, it will be extended to de novo design applications and other antibacterial targets, providing a versatile framework for antibiotics discovery
Wojtek Treyde