Identification of Novel Antibacterials Using Machine Learning Techniques

Area of Science:

• Computational chemistry and drug discovery
• Microbiology and infectious diseases
• Pharmaceutical sciences

Background:

• Pharmaceutical development of novel antibacterials faces high failure risks, despite an urgent need for agents against resistant strains.
• Existing in silico models are limited in scoring diverse molecules for antibacterial potency.
• A large, proprietary dataset of over 140,000 molecules with antibacterial activity against Escherichia coli was compiled.

Purpose of the Study:

• To develop an efficient in silico model for predicting antibacterial activity in novel chemical compounds.
• To identify promising drug candidates with high potential for antibacterial efficacy.
• To overcome limitations of current in silico methods for diverse molecular structures.

Main Methods:

• Application of six in silico techniques to analyze a dataset of 140,000+ molecules.
• External validation using 5,000 structurally diverse compounds.
• Utilized Kohonen-based nonlinear mapping, achieving the best predictive performance.

Main Results:

• The developed in silico model demonstrated an average predictive power of 75.5%.
• Several identified compounds exhibited significant antibacterial potency against Escherichia coli.
• Selected compounds were validated as inhibitors of bacterial translation machinery in vitro and in vivo.

Conclusions:

• The novel in silico model is effective for discovering potential antibacterial agents.
• Identified compounds show promise as new antibiotics, with favorable selectivity indices.
• Many discovered active compounds possess strong intellectual property positions.