Home Medizin Deep Learning erschließt neue Antibiotikaklassen und revolutioniert den Kampf gegen Resistenzen

Deep Learning erschließt neue Antibiotikaklassen und revolutioniert den Kampf gegen Resistenzen

von NFI Redaktion

According to a study published in Nature, a recent research used a Deep Learning approach to explore the chemical substructures contributing to the discovery of new classes of antibiotics.

Study: Discovering a new class of antibiotics using explainable Deep Learning. Image credit: Kasarp Studio / Shutterstock.com Study: Discovering a new class of antibiotics using explainable Deep Learning. Image credit: Kasarp Studio / Shutterstock.com

Discovery of New Antibiotic Classes

Prior studies have indicated a 38-year gap between the introduction of fluoroquinolones and oxazolidinones, highlighting the lengthy process of discovering new antibiotic classes. The ongoing antibiotic resistance crisis has underscored the importance of developing new antibiotics.

Resistance to antibiotics and the lack of new antibiotics have increased the morbidity from bacterial infections. Typically, antibiotics are discovered using a variety of approaches, including structure-guided and rational design, natural product mining, evolutionary and phylogenetic analyses, high-throughput screening, and in silico screens using machine learning.

Discovering novel antibiotic agents with broad structural diversity in the chemical space is highly challenging. To address this challenge, Deep Learning methods were employed to identify potential antibiotics from large chemical libraries.

For instance, Halicin and Abaucin were identified from the Drug Repurposing Hub, which contains over 6,000 molecules. This approach was also used to identify antibacterial agents from the ZINC15 library, which contains approximately 107 million molecules. The ZINC15 library utilized Chemprop, a platform for graphical neural networks, as opposed to typical Black-Box models or models that are not easily interpretable.

In the current study, graphical neural network models trained on large datasets linked to measurements of antibiotic activity and human cell cytotoxicity were employed. The hypothesis was that model predictions could be explained by chemical substructures identified using graph search algorithms.

As antibiotic classes are categorized based on shared substructures, the current study suggested that the identification of substructures for explaining model predictions could be utilized.

Models for Antibiotic Activity and Human Cell Cytotoxicity

The study sought to discover effective antibiotic classes against Staphylococcus aureus, a gram-positive pathogenic bacterium. This bacterium was selected due to its resistance to many first-choice antibiotics and its role in causing many difficult-to-treat healthcare-associated infections.

A total of 39,312 structurally distinct antibiotics were tested for antibiotic activity against a methicillin-sensitive strain of S. aureus. Approximately 1.3% of all compounds exhibited antibacterial activity.

Chemprop was used to train ensembles of graphical neural networks. The screening data were used for binary classification predictions, providing insights into whether a new compound can inhibit bacterial growth based on its chemical structure.

The graphical neural network implements complex steps based on the atoms and chemical bonds of each molecule. Through these complex steps, each model generated a prediction value between zero and one, indicating the likelihood of the molecule’s antibacterial activity.

Chemprop models with RDKit-computed molecular features, which exhibit molecular properties, can be used to predict antibiotic activity. The model outperformed other Deep Learning models, such as Random Forest.

Orthogonal models were used to predict cytotoxicity in human cells. The result was used to identify compounds that could be effective against S. aureus.

Some compounds were found to be cytotoxic to human liver carcinoma cells (HepG2), human lung fibroblast cells (IMR-90), and human primary skeletal muscle cells (HSkMCs). Cytotoxicity models for IMR-90 cells were more informative compared to HepG2 and HSkMCs.

Discovery of New Antibiotic Structure Classes

The current study identified putative antibiotic structure classes through diagram-based explanations of Deep Learning model predictions. These models were trained on the antibiotic activity and cytotoxicity of 12,076,365 compounds.

Several compounds exhibiting antibacterial activity against S. aureus were identified. However, one structure class showed superior selectivity and the ability to overcome resistance. Importantly, this class of antibiotics also exhibited favorable chemical and toxicological properties.

A mouse model demonstrated that the new structure class of antibiotics was effective in both topical and systemic treatment of methicillin-resistant S. aureus (MRSA) infection. Furthermore, structure-activity relationship analyses suggested that this structure class could be optimized for higher sensitivity and selectivity against gram-positive pathogens and improved permeability against gram-negative pathogens.

The current study underscored the effectiveness of the Deep Learning approach in discovering new antibiotic classes. A new structure class of antibiotics can be identified based on predictions of individual compound hits and the analysis of their chemical substructures. In addition to reducing the chemical space, another advantage of this approach is the ability to automate the identification of novel structural motifs.

A better understanding of graph-based explanation predictions could enable the discovery of new antibiotic classes. The current study’s approach could serve as a foundation for future predictions using Deep Learning models.

Journal Reference:

  • Wong, F., Zheng, E.J, Valeri, JA, et al. (2023) Discovering a new class of antibiotics using explainable Deep Learning. Nature. doi:10.1038/s41586-023-06887-8

Related Posts

Adblock Detected

Please support us by disabling your AdBlocker extension from your browsers for our website.