Background: Artificial intelligence as well as machine learning are now widely used, including in medicine. One potential use case is an otherwise lengthy and yet not necessarily successful process, the discovery of new antibiotics. This process, marked by setbacks and costly, is becoming less attractive to pharmaceutical companies, resulting in fewer antibiotics being developed. In addition, there is the increased incidence of multidrug-resistant germs that cannot be controlled by conventional antibiotics, creating additional demand.
Objective: This paper addresses the question of which machine learning methods are applied in which steps of the discovery of a new antibiotic, in which steps the application of the methods makes sense, which advantages and disadvantages they entail, and an outlook on further potentials.
Method: For the literature review, a forward search was performed on three different databases, starting from a search string containing keywords relevant to the topic. The initial 850 hits could be narrowed down to 25 relevant publications after further screening.
Results: The literature found shows that neural networks, support vector machines as well as decision trees have been used so far in the generation and discovery of structures of new potential drugs, but also in the assessment of potential efficacy.
Conclusion: In the future, ML methods are likely to be used more frequently, as an algorithm can work much faster in this use case, and the increasing computational capacity will likely accelerate over time. The focus is likely to be on KNN, as this algorithm is very easy to train, requires little preprocessing, and does not require kernels such as SVM.