Application of AI for Metallodrug Discovery

 Metal-based antibacterial agents offer key advantages over organic molecules: tunable chemistry, redox activity, and reduced toxicity. However, discovering these compounds is still largely a process of trial and error because we lack the datasets and predictive tools tailored for metal complexes. Recently, in collaboration with Prof. David Balcells from the University of Oslo,  our  group  developed  the  first  deep  learning  model  capable of  both  predicting  the anticancer  activity  and  de  novo  chemical  structure  of  metal  complexes (https://chemrxiv.org/doi/full/10.26434/chemrxiv-2025-pp32k/v2, currently under revision in Nat.  Biotech.).  In  this  project,  we  will  extend the  model's  scope  to  predict metal  complexes with  selective,  microbiome-modulating  properties.  We  will  benchmark  our  deep  learning model  against  standard  machine  learning  approaches  to  test  whether  metal-specific  training improves prediction of selective killing. Top predicted compounds will then undergo lab testing against pathogenic and beneficial bacteria, as well as mammalian cells. Within 12 months, the project  will  deliver:  1)  a  publicly  open  dataset  for  modeling  microbiome-modulating  and antibacterial  metallodrugs,  2)  a  rigorously  benchmarked  predictive  workflow,  and  3)experimentally  validated  candidates  to  support  future  funding  applications  and  translational studies.