KANAGAWA, Japan—Identifying ligand-binding sites on proteins is key to understanding how the protein functions and finding new drugs. Several algorithms have been developed to look for binding clefts in protein surfaces, but none of them into account a key identifier of protein-ligand binding, the clustering of specific amino acids. To address this problem, researchers at Astellas Pharma and Tokai University developed a new algorithm that allowed them to predict drug and druglike molecule binding.
As they report in the Journal of Chemical Information and Modeling, the researchers started with X-ray structures of proteins bound by drugs or druglike ligands, the latter defined by 14 molecular descriptors. Using the Alpha Site Finder function of MOE, they identified concavities in the proteins' surfaces and characterized the composition of amino acids in these concavities, as well as more general placement on protein surfaces. They found specific amino acids tended to be found within concavities more often than others, and they used this information to develop an index called propensity for ligand binding (PLB).
The researchers then tested their PLB index on a collection of 756 proteins, identifying 15,232 potential binding sites. When they looked at the highest scoring concavities on a given protein, they found 79% represented known ligand binding sites. When they extended that to the two highest scoring sites, the success rate was 86%.
"The method may be applicable to relatively low-resolution X-ray structures and those constructed using homology modeling," say the authors. "The PLB index would also be useful for identifying ligand-binding sites on novel target molecules with unknown ligands."