Computational models, databases and tools for antibiotic combinations

Antibiotic combinations prediction based on machine learning to multicentre clinical data and drug interaction correlation

BACKGROUND

With increasing antibiotic resistance and regulation, the issue of antibiotic combination has been emphasised. However, antibiotic combination prescribing lacks a rapid identification of feasibility, while its risk of drug interactions is unclear.

METHODS

We conducted statistical descriptions on 16 101 antibiotic coprescriptions for inpatients with bacterial infections from 2015 to 2023. By integrating the frequency and effectiveness of prescriptions, we formulated recommendations for the feasibility of antibiotic combinations. Initially, a machine learning algorithm was utilised to optimise grading thresholds and habits for antibiotic combinations. A feedforward neural network (FNN) algorithm was employed to develop antibiotic combination recommendation model (ACRM). To enhance interpretability, we combined sequential methods and DrugBank to explore the correlation between antibiotic combinations and drug interactions.

RESULTS

A total of 55 antibiotics, covering 657 empirical clinical antibiotic combinations were used for ACRM construction. Model performance on the test dataset showed AUROCs of 0.589-0.895 for various antibiotic recommendation classes. The ACRM showed satisfactory clinical relevance with 61.54-73.33% prediction accuracy in a new independent retrospective cohort. Antibiotic interaction detection showed that the risk of drug interactions was 29.2% for strongly recommended and 43.5% for not recommended. A positive correlation was identified between the level of clinical recommendation and the risk of drug interactions.

CONCLUSIONS

Machine learning modelling of retrospective antibiotic prescriptions habits has the potential to predict antibiotic combination recommendations. The ACRM plays a supporting role in reducing the incidence of drug interactions. Clinicians are encouraged to adopt such systems to improve the management of antibiotic usage and medication safety.

AADB: A Manually Collected Database for Combinations of Antibiotics With Adjuvants

Antimicrobial resistance is a global public health concern. The lack of innovations in antibiotic development has led to renewed interest in antibiotic adjuvants. However, there is no database to collect antibiotic adjuvants. Herein, we build a comprehensive database named Antibiotic Adjuvant DataBase (AADB) by manually collecting relevant literature. Specifically, AADB includes 3,035 combinations of antibiotics with adjuvants, covering 83 antibiotics, 226 adjuvants, and 325 bacterial strains. AADB provides user-friendly interfaces for searching and downloading. Users can easily obtain these datasets for further analysis. In addition, we also collected related datasets (e.g., chemogenomic and metabolomic data) and proposed a computational strategy to dissect these datasets. As a test case, we identified 10 candidates for minocycline, and 6 of 10 candidates are the known adjuvants that synergize with minocycline to inhibit the growth of E. coli BW25113. We hope that AADB can help users to identify effective antibiotic adjuvants. AADB is freely available at http://www.acdb.plus/AADB.

Identification of Robust Antibiotic Subgroups by Integrating Multi-Species Drug-Drug Interactions

Previous studies have shown that antibiotics can be divided into groups, and drug-drug interactions (DDI) depend on their groups. However, these studies focused on a specific bacteria strain (i.e., Escherichia coli BW25113). Existing datasets often contain noise. Noisy labeled data may have a bad effect on the clustering results. To address this problem, we developed a multi-source information fusion method for integrating DDI information from multiple bacterial strains. Specifically, we calculated drug similarities based on the DDI network of each bacterial strain and then fused these drug similarity matrices to obtain a new fused similarity matrix. The fused similarity matrix was combined with the T-distributed stochastic neighbor embedding algorithm, and hierarchical clustering algorithm can effectively identify antibiotic subgroups. These antibiotic subgroups are strongly correlated with known antibiotic classifications, and group-group interactions are almost monochromatic. In summary, our method provides a promising framework for understanding the mechanism of action of antibiotics and exploring multi-species group-group interactions.