Aim 3

Generate structural predictions of target binding to identify and optimize antibiotic leads

One of the most common approaches in antibiotic discovery is to screen for compounds that bind to specific, highly relevant bacterial targets (typically an important protein). To pursue this, we will use DNA-encoded compound libraries (DELs), which allow for simultaneous testing of the binding properties of millions of compounds at once, against a given bacterial target. This approach uses DNA barcoding to tag different compounds with unique DNA tags. These DNA-tagged compound libraries can then be applied to the relevant bacterial target purified from bacterial cells. Compounds that specifically bind to the target of interest can subsequently be identified using their DNA tags, allowing millions of compounds to be screened at once.

To design our DNA-encoded compound libraries, we will rely on predictions from the generative drug design models to narrow the search to relevant areas of chemical space for binding to a given target, based on the target’s 3D structure. Experimental data on the binding affinities of these in silico drug predictions to the bacterial target will then be used to further train the models, to improve their subsequent predictions.

As our models progressively develop abstract representations of the 3D structures of compounds and targets and their binding affinities, they may also be used to optimize the structure-activity relationships of promising leads further downstream. This process relies on exploring minor chemical modifications of the lead compound to optimize its efficacy, pharmacokinetics, toxicity profile, etc, to exploit the chemical space surrounding an effective drug design.

Image: Graphical overview of the DNA-encoded compound library screening approach for identifying compounds that bind to a specific bacterial target.