Steven Ramsey’s research while affiliated with Lehman College and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (15)


Solvation Thermodynamic Costs of Cognate Binding Site Formation
  • Preprint

January 2025

·

4 Reads

Yeonji Ji

·

Vjay Molino

·

Steven Ramsey

·

Conformational flexibility complicates the identification of lead molecules that are shape and charge complementary to target proteins. Solvation thermodynamics has typically not been integrated into the exploration of alternate protein conformations. Here, we study the variation of solvation thermodynamic potentials as proteins adopt different conformations. Specifically, we analyze solvation thermodynamics of protein binding cavities with conformations obtained from molecular dynamics simulations with mobile side chains and side chains restrained about their cognate bound structure. We find that the reorganization of protein side chains has a significant effect on the structure and thermodynamics of binding site solvation and, in the vast majority of cases, that there is a significant solvation free energetic cost to forming cognate ligand bound structures when the ligand is absent. We discuss how understanding the interplay between solvation thermodynamics and protein structural fluctuations is crucial for discovering alternative binding pockets, estimating the contribution to binding affinity of displacing water upon ligand binding, and assessing revealed cryptic pocket bindability.


A Self Consistent Approach to Rotamer and Protonation State Assignments (RAPA): Moving Beyond Single Protein Configurations

December 2024

·

4 Reads

Mossa Ghattas

·

Prerna Gera

·

Steven Ramsey

·

[...]

·

There are currently over 160,000 protein crystal structures obtained by X-ray diffraction with resolutions of 1.5Å or greater in the Protein Data Bank. At these resolutions hydrogen atoms do not resolve and heavy atoms such as oxygen, carbon, and nitrogen are indistinguishable. This leads to ambiguity in the rotamer and protonation states of multiple amino acids, notably asparagine, glutamine, histidine, serine, tyrosine, and threonine. When the rotamer and protonation states of these residues changes, so too does the electrochemical surface of a binding site. A variety of computational tools have been developed to assign these states for these residues based on a crystal protein structure by evaluating the possible states and typically deciding on one single state for each residue. We posit that multiple rotamer and protonation states of residues are consistent with the resolved structure of the proteins and introduce a protonation and rotamer assignment tool that identifies an ensemble of rotamer and protonation states that are consistent with the X-ray scattering data of the protein. Here, we present a Rotamer and Protonation state Assignment (RAPA) tool that analyzes local hydrogen bonding environments in the resolved structures of proteins and identifies a set of unique rotamer and protonation states that are energetically consistent with the crystal structure. We evaluate all RAPA predicted states in unrestrained molecular dynamics simulations and find that there are multiple configurations for each protein which match the X-ray results with RMSDs of less than 1.0Å for the atoms with the lowest 90% B-factors. We find that for most protein systems (62 of 77) there are 8 or fewer possible states suggesting that there is no combinatorial explosion of accessible configurations for a majority of proteins. This suggests that investigating all energetically accessible rotamer and protonation states for most proteins is computationally feasible and that the selection of single states is arbitrary.


Solvation Thermodynamic Costs of forming Cognate Binding Site Formation

December 2024

·

7 Reads

Proteins are inherently flexible which complicates the identification of lead molecules that are shape and charge complementary to target proteins. While significant effort has been dedicated to exploring alternate protein conformations, solvation thermodynamics has typically not been integrated into these studies. Here, we study how solvation thermodynamics fluctuate as proteins adopt different conformations. We analyze solvation thermodynamics within the binding cavities of conformations for which side chains are mobile in molecular dynamics simulations and compare these to conformations for which they remain restrained about the cognate bound structure. We identify structural motifs that present significant costs to the sampling of cognate ligand bound structures. We find that the reorganization of protein side chains has a significant effect on the structure and thermodynamics of binding site solvation. We discuss how understanding the interplay between solvation thermodynamics and protein structural fluctuations is crucial for both discovering alternative binding pockets, estimating the contribution to binding affinity of displacing water upon ligand binding, and assessing revealed cryptic pocket bindability.


Crystal structure of the dopamine D3 receptor with eticlopride bound at the binding site
[28] This representation shows the approximate position of the orthosteric binding site (OBS) with a blue oval and the secondary binding site (SBS) with a green oval.
Structure of the (-)-stepholidine core with four rings annotated alphabetically as referenced in the text
R1 represents the substitution at the C3 position. The chiral carbon is labeled by a star.
Hydration sites and corresponding AGBNP2 spheres at the dopamine D3 receptor binding site
(a) Location of hydration sites (red) within the binding cavity of the Dopamine D3 receptor as mapped by Hydration Site Analysis. (b) Hydration spheres (green) of the AGBNP2 model for the same receptor structure in (a). The positions of the AGBNP2 hydration spheres are functions of the internal coordinates of the receptor.
Strategy for scoring and placement of AGBNP2 hydration spheres in dopamine D3 receptor binding site
(a) Location of a hydration site identified by HSA using three receptor structures (residues from one receptor structure shown for clarity); the overlapping red, yellow and orange spheres represent a hydration site identified by each receptor structure; the energetic penalties incurred from each HSA map are annotated in kcal/mol, (b) An AGBNP2 hydration sphere (green) is placed and scored by averaging the energetic penalties from the three maps at the location of the HSA site; the AGBNP2 hydration sphere is placed at the geometrical center of the atoms represented in CPK and is anchored to respective atoms during the simulation.
Interactions of C3 pentyl analogue with the dopamine D3 receptor
a) The C3 pentyl analogue (3e, purple) of (-)-stepholidine is observed to interact with Ser 192 of the receptor at the orthosteric binding site. In order for the C3 analogues to interact with Ser 192, the C10 hydroxyl group is placed in proximity of Ser 192; b) The 3e C3 analogue in another observed binding pose in which it interacts with Ser 196, rather than Ser 192. In this pose, ring D of the (-)-stepholidine core is bound deeper into the orthosteric binding site and the ligand is twisted causing Tyr 365 in the SBS to rotate and move away from Ser 182 of ECL2. The receptor is represented as a pink ribbon.

+5

Inclusion of enclosed hydration effects in the binding free energy estimation of dopamine D3 receptor complexes
  • Article
  • Full-text available

September 2019

·

131 Reads

·

9 Citations

Confined hydration and conformational flexibility are some of the challenges encountered for the rational design of selective antagonists of G-protein coupled receptors. We present a set of C3-substituted (-)-stepholidine derivatives as potent binders of the dopamine D3 receptor. The compounds are characterized biochemically, as well as by computer modeling using a novel molecular dynamics-based alchemical binding free energy approach which incorporates the effect of the displacement of enclosed water molecules from the binding site. The free energy of displacement of specific hydration sites is obtained using the Hydration Site Analysis method with explicit solvation. This work underscores the critical role of confined hydration and conformational reorganization in the molecular recognition mechanism of dopamine receptors and illustrates the potential of binding free energy models to represent these key phenomena.

Download

Hidden bias in the DUD-E dataset leads to misleading performance of deep learning in structure-based virtual screening

August 2019

·

711 Reads

·

238 Citations

Recently much effort has been invested in using convolutional neural network (CNN) models trained on 3D structural images of protein-ligand complexes to distinguish binding from non-binding ligands for virtual screening. However, the dearth of reliable protein-ligand x-ray structures and binding affinity data has required the use of constructed datasets for the training and evaluation of CNN molecular recognition models. Here, we outline various sources of bias in one such widely-used dataset, the Directory of Useful Decoys: Enhanced (DUD-E). We have constructed and performed tests to investigate whether CNN models developed using DUD-E are properly learning the underlying physics of molecular recognition, as intended, or are instead learning biases inherent in the dataset itself. We find that superior enrichment efficiency in CNN models can be attributed to the analogue and decoy bias hidden in the DUD-E dataset rather than successful generalization of the pattern of protein-ligand interactions. Comparing additional deep learning models trained on PDBbind datasets, we found that their enrichment performances using DUD-E are not superior to the performance of the docking program AutoDock Vina. Together, these results suggest that biases that could be present in constructed datasets should be thoroughly evaluated before applying them to machine learning based methodology development.


Fig 3. The performance of receptor-ligand and receptor-ligand-water CNN models 273 in 10 DUD-E targets. 274
Fig 5. Performance of the receptor-ligand model for the same ligand test sets with 339
Fig 8. Inter-target prediction performance of ligand-only CNN models A) tested on test 501
Hidden Bias in the DUD-E Dataset Leads to Misleading Performance of Deep Learning in Structure-Based Virtual Screening

March 2019

·

25 Reads

·

13 Citations

p>Recently much effort has been invested in using convolutional neural network (CNN) models trained on 3D structural images of protein-ligand complexes to distinguish binding from non-binding ligands for virtual screening. However, the dearth of reliable protein-ligand x-ray structures and binding affinity data has required the use of constructed datasets for the training and evaluation of CNN molecular recognition models. Here, we outline various sources of bias in one such widely-used dataset, the Directory of Useful Decoys: Enhanced (DUD-E). We have constructed and performed tests to investigate whether CNN models developed using DUD-E are properly learning the underlying physics of molecular recognition, as intended, or are instead learning biases inherent in the dataset itself. We find that superior enrichment efficiency in CNN models can be attributed to the analogue and decoy bias hidden in the DUD-E dataset rather than successful generalization of the pattern of protein-ligand interactions. Comparing additional deep learning models trained on PDBbind datasets, we found that their enrichment performances using DUD-E are not superior to the performance of the docking program AutoDock Vina. Together, these results suggest that biases that could be present in constructed datasets should be thoroughly evaluated before applying them to machine learning based methodology development. </p


Fig 3. The performance of receptor-ligand and receptor-ligand-water CNN models 273 in 10 DUD-E targets. 274
Fig 5. Performance of the receptor-ligand model for the same ligand test sets with 339
Fig 8. Inter-target prediction performance of ligand-only CNN models A) tested on test 501
Hidden bias in the DUD-E dataset leads to misleading performance of deep learning in 2 structure-based virtual screening 3 4

March 2019

·

230 Reads

24 25 Recently much effort has been invested in using convolutional neural network (CNN) 26 models trained on 3D structural images of protein-ligand complexes to distinguish binding 27 from non-binding ligands for virtual screening. However, the dearth of reliable protein-28 ligand x-ray structures and binding affinity data has required the use of constructed 29 datasets for the training and evaluation of CNN molecular recognition models. Here, we 30 outline various sources of bias in one such widely-used dataset, the Directory of Useful 31 Decoys: Enhanced (DUD-E). We have constructed and performed tests to investigate 32 whether CNN models developed using DUD-E are properly learning the underlying 33 physics of molecular recognition, as intended, or are instead learning biases inherent in 34 the dataset itself. We find that superior enrichment efficiency in CNN models can be 35 attributed to the analogue and decoy bias hidden in the DUD-E dataset rather than 36 successful generalization of the pattern of protein-ligand interactions. Comparing 37 additional deep learning models trained on PDBbind datasets, we found that their 38 enrichment performances using DUD-E are not superior to the performance of the 39 docking program AutoDock Vina. Together, these results suggest that biases that could 40 be present in constructed datasets should be thoroughly evaluated before applying them 41 to machine learning based methodology development. 42 43 44 45 46 3


New Dopamine D3-Selective Receptor Ligands Containing a 6-Methoxy-1,2,3,4-tetrahydroisoquinolin-7-ol Motif

September 2018

·

75 Reads

·

11 Citations

ACS Medicinal Chemistry Letters

A series of analogues featuring a 6-methoxy-1,2,3,4-tetrahydroisoquinolin-7-ol unit as the arylamine “head” group of a classical D3 antagonist core structure, were synthesized and evaluated for affinity at dopamine D1, D2 and D3 receptors (D1R, D2R, D3R). The compounds generally displayed strong affinity for D3R with very good D3R selectivity. Docking studies at D2R and D3R crystal structures revealed that the molecules are oriented such that their arylamine units are positioned in the orthosteric binding pocket of D3R, with the arylamide “tail” units residing in the secondary binding pocket. Hydrogen bonding between Ser 182 and Tyr 365 at D3R stabilize extracellular loop 2 (ECL2), which in turn con-tributes to ligand binding by interacting with the "tail" units of the ligands in the secondary binding pocket. Similar interac-tions between ECL2 and the “tail” units were absent at D2R due to different positioning of the D2R loop region. The pres-ence of multiple H-bonds with the phenol moiety of the head group of 7 and Ser192 accounts for its stronger D3R affinity as compared to the 6,7-dimethoxy-1,2,3,4-tetrahydroisoquinoline-containing analogue 8.



Water Pharmacophore: Designing Ligands using Molecular Dynamics Simulations with Water

July 2018

·

560 Reads

·

30 Citations

In this study, we demonstrate a method to construct a water-based pharmacophore model which can be utilized in the absence of known ligands. This method utilizes waters found in the binding pocket, sampled through molecular dynamics. Screening of compound databases against this water-based pharmacophore model reveals that this approach can successfully identify known binders to a target protein. The method was tested by enrichment studies of 7 therapeutically important targets and compared favourably to screening-by-docking with Glide. Our results suggest that even without experimentally known binders, pharmacophore models can be generated using molecular dynamics with waters and used for virtual screening.


Citations (9)


... Based on previous molecular modeling results, this behavior was attributed to the ability of the latter group of compounds to form two hydrogen bond interactions between the primary pharmacophore group and Ser192 of the receptor compared to only one hydrogen bond in the case of the 6,7dimethoxy-1,2,3,4-tetrahydroisoquinoline-containing compounds. 19,25 A similar molecular docking study conducted here indicates that rigidification of the linker induces a mode of binding distinct from the one observed earlier and consistent with the observed SAR. As shown in Fig. 3, the three most potent compounds (5s, 5t, and 6a) dock as generally expected with the primary pharmacophore group in the orthosteric pocket and forming the key salt bridge between the alkyl protonated nitrogen and Asp110. ...

Reference:

New tetrahydroisoquinoline-based D3R ligands with an o-xylenyl linker motif
Inclusion of enclosed hydration effects in the binding free energy estimation of dopamine D3 receptor complexes

... One of the main challenges associated with the database are issues arising from the limited size or diversity of the prepared data set which leads to limited generalizability and biased results. [16,138] As such, the bigger and more diverse a dataset is the better the expected outcome of a VS study. Increasing the data set size and diversity will also solve the hit rate issues and biased results commonly observed with small or diversly biased data sets [9,16,136,137]. ...

Hidden bias in the DUD-E dataset leads to misleading performance of deep learning in structure-based virtual screening

... D 2 like-selective compounds showed binding affinities ≤500 nM at D 2 R and D 3 R. The final datasets consisting of 29 (SC1-SC29 [46-60]), 25 (SC30-SC54 [53,[61][62][63][64][65][66][67][68][69][70][71][72]), 152 (SC55-SC206 [56,69,) and 78 (SC207-SC284 [53,56,62,63,65,66,79,87,89,96,[99][100][101][103][104][105][106][107][108][109][110][111][112][113][114][115][116][117][118]) molecules, respectively, are shown in Tables S1-S4. ...

New Dopamine D3-Selective Receptor Ligands Containing a 6-Methoxy-1,2,3,4-tetrahydroisoquinolin-7-ol Motif
  • Citing Article
  • September 2018

ACS Medicinal Chemistry Letters

... The results of alanine scanning indicated that Arg3, Gly4, HxG5, and Trp6 strongly contributed to the binding activity. The GIST analysis and molecular dynamics simulation suggested that entropically unfavorable waters were clustered in the cryptic pocket, and the replacement of unstable water molecules by the two lipophilic side chains of HxG5 and Trp6 contributed to the strong binding upon the target 64,65 . Moreover, GIST indicated that the entropically unfavored water molecules frequently stayed around the carbonyl groups of Gly4 and HxG5 of macrocyclic peptide 1 (1), and the team speculated that these two hydrogen bond acceptors (HBAs) would act as anchoring interactions that displace the restrained entropically unfavored water molecules and form electrostatic interactions with NNMT. ...

Water Pharmacophore: Designing Ligands using Molecular Dynamics Simulations with Water

... The water densities were calculated by considering a grid-based spatial decomposition, which allowed for the identification of high-density regions around the ligand. Interaction energies were computed to assess the strength and stability of water-ligand interactions throughout the trajectory [44][45][46]. The occupancy frequency of water molecules around the protein-ligand complex was monitored across the entire trajectory. ...

Solvation Structure and Thermodynamic Mapping (SSTMap): An Open-Source, Flexible Package for the Analysis of Water in Molecular Dynamics Trajectories
  • Citing Article
  • November 2017

Journal of Chemical Theory and Computation

... So far, our attempts to identify new selective D 1 R partial agonist/D 3 R antagonist dual-acting ligands from THPBs have not been successful, although we have identified several dopamine receptor subtype-specific ligands (e.g., 77− 80). [182][183][184][185]187 Our structure−affinity relationship studies on L-SPD revealed that bromination at the C-12 position improved D 1 R affinity but generally diminished D 3 R affinity. Homologation at the C-9 position improved D 1 R selectivity versus D 2 R and D 3 R (often lacking affinity at D 2 R), whereas alkylation of the C-10 phenolic group with up to 6 carbon atoms in length was tolerated for D 1 R affinity. ...

Synthesis and evaluation of C9 alkoxy analogues of (-)-stepholidine as dopamine receptor ligands
  • Citing Article
  • September 2016

European Journal of Medicinal Chemistry

... These grids provide a visualization of the hydration patterns about the protein binding site. 38 Accounting for receptor desolvation during docking with GIST has been implemented in DOCK 3. 31,32 Many other methods have been used to account for solvation effects in molecular docking. 39−41 As an alternative to GIST, 3D-RISM can be used to generate solvation energy grids that can be readily incorporated into receptor desolvation scoring. ...

Solvation thermodynamic mapping of molecular surfaces in AmberTools: GIST
  • Citing Article
  • June 2016

Journal of Computational Chemistry

... In the case of interaction with proteins, the structural nature of the three-dimensional network of water involved in their hydration results from many factors, including the chemical composition of the amino acid surface constituting any polypeptide (Haider et al., 2016). The stability of intermolecular interactions between water and proteins mainly involves acid, basic and hydrophilic side chain amino acids. ...

Enthalpic Breakdown of Water Structure on Protein Active-Site Surfaces
  • Citing Article
  • May 2016

The Journal of Physical Chemistry B

... So far, our attempts to identify new selective D 1 R partial agonist/D 3 R antagonist dual-acting ligands from THPBs have not been successful, although we have identified several dopamine receptor subtype-specific ligands (e.g., 77− 80). [182][183][184][185]187 Our structure−affinity relationship studies on L-SPD revealed that bromination at the C-12 position improved D 1 R affinity but generally diminished D 3 R affinity. Homologation at the C-9 position improved D 1 R selectivity versus D 2 R and D 3 R (often lacking affinity at D 2 R), whereas alkylation of the C-10 phenolic group with up to 6 carbon atoms in length was tolerated for D 1 R affinity. ...

Tetrahydroprotoberberine alkaloids with dopamine and σ receptor affinity
  • Citing Article
  • March 2016

Bioorganic & Medicinal Chemistry