Predicting ligand binding to proteins by affinity fingerprinting

Area of Science:

• Biochemistry
• Computational Chemistry
• Drug Discovery

Background:

• Traditional methods for predicting molecular properties rely on physical representations like chemical structures or spectra.
• Predicting biological activity often depends on these established molecular representations.
• A novel approach uses compound binding potency against diverse proteins to create 'affinity fingerprints'.

Purpose of the Study:

• To introduce and validate the concept of affinity fingerprints for molecular representation.
• To demonstrate the utility of affinity fingerprints in predicting compound binding properties against new proteins.
• To explore shared binding interaction patterns across diverse proteins.

Main Methods:

• Generated affinity fingerprints for 122 diverse compounds using a reference panel of eight proteins.
• Applied multivariate regression to create computational surrogates for new proteins.
• Validated the predictive power of these surrogates on various enzymes and a large compound library.

Main Results:

• Affinity fingerprints uniquely identified about 75% of tested small organic compounds.
• Computational surrogates effectively predicted binding potencies for new proteins, even those structurally dissimilar to the reference panel.
• The method successfully predicted several sub-micromolar hits in a library of 5000 compounds.

Conclusions:

• Affinity fingerprint databases offer insights into protein similarities and can challenge existing structural homology theories.
• This approach facilitates efficient pre-screening of large compound libraries for drug design.
• Combining fingerprint similarities with minimal empirical testing aids in selecting promising drug candidates.