Predictive cartography of metal binders using generative topographic mapping.

Area of Science:

• Computational chemistry
• Cheminformatics
• Materials science

Background:

• Understanding metal-ligand interactions is crucial for designing selective binders.
• Predicting binding affinities and selectivity for diverse metal ions remains a challenge.
• Generative topographic mapping (GTM) offers a novel approach for chemical space visualization.

Purpose of the Study:

• To visualize the chemical space of organic molecules concerning their binding to 41 metal cations.
• To develop predictive models for the stability constants (logK) of 1:1 metal-ligand complexes.
• To explore the utility of GTM for metal binder design and chemical space exploration.

Main Methods:

• Application of Generative Topographic Mapping (GTM) for data visualization and model building.
• Utilizing "density maps," "activity landscapes," and "selectivity landscapes" for analysis.
• Cross-validation procedures to assess the predictive performance of GTM-based models.

Main Results:

• A 2D map of 2962 metal binders revealed selectivity and promiscuity zones for various metals.
• GTM-based models demonstrated good predictive performance for both global and local datasets.
• Data likelihood was identified as a metric for defining the applicability domain of GTM models.

Conclusions:

• GTM is an effective tool for the predictive cartography of metal binders.
• The approach enables visualization of chemical space and prediction of metal affinity profiles for new ligands.
• This method facilitates the discovery and design of novel metal-binding agents.