This proposal describes efforts to develop, test, and improve a machine learning approach to identify potential
metallo-ß-lactamase (MBL) inhibitors from large chemical libraries. The MBLs are bacterial enzymes that are
becoming more prevalent in the clinic and are leading to more incidents of antibiotic resistance in once easily-
treatable infections, including secondary infections in COVID patients. There have been tremendous efforts
to identify new MBL inhibitors; however to date, there are no clinical inhibitors of these enzymes. This proposal
describes a novel, multidisciplinary approach to identify new MBL inhibitors.
In Specific Aim 1, we propose to improve our initial computer model, which currently ranks compounds in
chemical libraries on their likelihood of being a potential MBL inhibitor, based on previous inhibition data
collected on MBL NDM-1. The improved model will be developed using data sets containing inhibition data
from quantitative HTS (qHTS) experiments on MBLs, NDM-1, VIM-2, and IMP-1. The new model will be used
to screen the 1.3 million compound-containing ChemBridge chemical library, and qHTS studies with VIM-2,
IMP-1, and NDM-1 will be conducted to test the results from the virtual screenings. The final, validated model
will be made available to the public on our MBL inhibitor website. In Specific Aim 2, we will perform
microbiological, structural, and biochemical studies on the top 500 compounds identified by our model and
confirmed with qHTS. In addition to minimal inhibitor concentration values, we propose to determine the
mechanism of inhibition and to structurally-characterize the enzyme-inhibitor complexes.
It is hoped that this machine learning approach will identify novel, pan MBL inhibitors and compounds that
can be further re-designed. In addition, it is hoped that this approach, once developed and tested, can be
used to identify inhibitors of other biomedically-important enzymes.