Combinatorial discovery of antibacterials via a feature-fusion based machine learning workflow

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Combating drug-resistant bacteria, such as methicillin-resistant Staphylococcus aureus (MRSA), requires the discovery of novel antibacterials.
• Traditional screening methods are time-consuming and laborious.
• Existing machine learning approaches are limited by the availability of reliable datasets.

Purpose of the Study:

• To develop an efficient workflow for identifying novel antibacterial compounds.
• To address the limitations of traditional screening and current machine learning methods in antibacterial discovery.
• To identify lead compounds effective against MRSA and capable of overcoming existing resistance mechanisms.

Main Methods:

• A combinatorial library of over 110,000 candidates was synthesized using the Ugi reaction.
• A focused library was generated via uniform sampling to reduce screening scale.
• A novel latent space constraint neural network, integrating fingerprint and physicochemical descriptors, was developed for predicting antibacterial properties.

Main Results:

• The developed machine learning model accurately predicted antibacterial properties by leveraging fused molecular descriptors.
• Three lead compounds exhibiting excellent efficacy against MRSA were identified.
• The identified compounds demonstrated potential in alleviating existing drug resistance.

Conclusions:

• The integration of machine learning with combinatorial chemistry accelerates high-quality data generation and mining for antibacterial screening.
• This workflow offers a promising strategy for the efficient discovery of novel antibacterials.
• The identified lead compounds represent a significant advancement in combating MRSA infections.