PoLi: A Virtual Screening Pipeline Based on Template Pocket and Ligand Similarity

Abstract

Often in pharmaceutical research the goal is to identify small molecules that can interact with and appropriately modify the biological behavior of a new protein target. Unfortunately, most proteins lack both known structures and small molecule binders, prerequisites of many virtual screening, VS, approaches. For such proteins, ligand homology modeling, LHM, that copies ligands from homologous and perhaps evolutionarily distant template proteins, has been shown to be a powerful VS approach to identify possible binding ligands. However, if we want to target a specific pocket for which there is no homologous holo template protein structure, then LHM will not work. To address this issue, in a new pocket-based approach, PoLi, we generalize LHM by exploiting the fact that the number of distinct small molecule ligand-binding pockets in proteins is small. PoLi identifies similar ligand-binding pockets in a holo template protein library, selectively copies relevant parts of template ligands, and uses them for VS. In practice, PoLi is a hybrid structure and ligand-based VS algorithm that integrates 2D fingerprint-based and 3D shape-based similarity metrics for improved virtual screening performance. On standard DUD and DUD-E benchmark databases, using modeled receptor structures, PoLi achieves an average enrichment factor of 13.4 and 9.6, respectively, in the top 1% of the screened library. In contrast, traditional docking-based VS using AutoDock Vina and homology-based VS using FINDSITE(filt) have an average enrichment of 1.6 (3.0) and 9.0 (7.9) on the DUD (DUD-E) sets, respectively. Experimental validation of PoLi predictions on dihydrofolate reductase, DHFR, using differential scanning fluorimetry, DSF, identifies multiple ligands with diverse molecular scaffolds, thus demonstrating the advantage of PoLi over current state-of-the-art VS methods.

Figure 1

Schematic flowchart of the PoLi virtual screening pipeline.

Figure 2

Comparison of a pocket based method (APoc) with global structure alignment and homology based approaches to detect similar ligands. The benchmark shows the ability of different approaches to recognize 30,000 pairs of similar ligands from 35,000 pairs of chemically dissimilar ligands.

Figure 3

Structure quality and binding site prediction accuracy for DUD and DUD-E proteins. Box and whiskers plot of (A) TM-score and (B) ligand binding pocket C_α RMSD of TASSER models to the experimentally determined structures. (C) Distance between the geometric center of the ligand in the co-crystallized complex and the center of the best predicted ligand-binding pocket in the 40 DUD and 65 DUD-E protein targets.

Figure 4

Thermal unfolding curves of E. coli DHFR A) Primary unfolding curves for hits belonging to the 1,3,5 triazine-2, 4-diamine group B) Primary unfolding curves for hits belonging to the quinazoline-1, 3-diamine group C) Primary unfolding curves for hits belonging to the pyrimidinediamine and aminopteridine group D) Primary unfolding curves for hits belonging to chemical classes distinct from any reported DHFR inhibitors E) Gaussian fit of first-derivative for curves in (A) F) Gaussian fit of first-derivative for curves in (B) G) Gaussian fit of first-derivative for curves in (C) H) Gaussian fit of first-derivative for curves in (D). On the plots A-D, the y-axis represents the normalized fluorescence and the x-axis represents the temperature in degrees Celsius. The experimental data points were fit to the respective equations using the nonlinear curve-fitting algorithm of GraphPad Prism v 6.0e.

Figure 5

Structures of small molecules showing binding to E. coli DHFR as assessed by the thermal shift assay methodology A) 1,3,5-triazine-2,4-diamine derivatives B) quinazoline-1,3-diamine derivatives C) Pyrimidinediamine and diaminopetridine derivatives D) 2,4 dihydroxyphenyl derivatives. The SDF files for the small molecules were downloaded from Pubchem (http://pubchem.ncbi.nlm.nih.gov) and the figure was generated using ChemBioDraw 14.0.