Essential papers: Difference between revisions
m (Added to-do topics.) |
(PAINS paper, titles of some papers, additional linebreaks for readability) |
||
| (10 intermediate revisions by 2 users not shown) | |||
| Line 3: | Line 3: | ||
== On using DOCK for virtual screening == | == On using DOCK for virtual screening == | ||
''' | '''Virtual screening of chemical libraries''' -- [https://doi.org/10.1038/nature03197 Nature (2004)] | ||
Commentary written by BKS over two decades ago. Even to this day, we face the same challenges and celebrate the same successes outlined in this paper. The ideas covered here have become recurrent themes in our lab, and paint with broad strokes a research enterprise that has deviated remarkably little. | |||
'''A practical guide to large-scale docking''' -- [https://doi.org/10.1038/s41596-021-00597-z Nature Protocols (2021)] | '''A practical guide to large-scale docking''' -- [https://doi.org/10.1038/s41596-021-00597-z Nature Protocols (2021)] | ||
Discussion of general purpose concerns with regards to virtual screening, and guidelines for ensuring adequate control calculations. An incredibly useful resource. See also the Supplementary Information, which provides explanation for each row found in an INDOCK file (one of the core input files for DOCK 3.x). | Discussion of general purpose concerns with regards to virtual screening, and guidelines for ensuring adequate control calculations. An incredibly useful resource. See also the Supplementary Information, which provides explanation for each row found in an INDOCK file (one of the core input files for DOCK 3.x). | ||
'''A paper on ZINC, whether the original, ZINC16, ZINC20, ZINC22, or another''' -- tbd which one | '''A paper on ZINC, whether the original, ZINC16, ZINC20, ZINC22, or another''' -- tbd which one | ||
'''The impact of library size and scale on virtual screening''' -- [https://doi.org/10.1038/s41589-024-01797-w Nature Chem Bio (2025)] | '''The impact of library size and scale on virtual screening''' -- [https://doi.org/10.1038/s41589-024-01797-w Nature Chem Bio (2025)] | ||
Bigger is better, and this paper demonstrates that screening larger libraries increases the likelihood of finding high-affinity hits. Just as importantly, this paper looks at how many compounds must be purchased/tested in order to reliably determine the hit rate from a given screen. | Bigger is better, and this paper demonstrates that screening larger libraries increases the likelihood of finding high-affinity hits. Just as importantly, this paper looks at how many compounds must be purchased/tested in order to reliably determine the hit rate from a given screen. | ||
== On how | |||
'''Read the dock docs''' -- [https://docs.docking.org/index.html Available here] | |||
The dock documentation contains the minimal instructions needed to run DOCK on our clusters, including commands that you can copy/paste in your own work. Functionally, our standard operating procedures for docking. | |||
== On benchmark sets == | |||
'''Directory of Useful Decoys, Enhanced (DUD-E): Better Ligands and Decoys for Better Benchmarking ''' -- [https://doi.org/10.1021/jm300687e J Med Chem (2012)] | |||
A rich set of ligands and decoys across 102 target proteins, spanning GPCRs, ion channels, kinases and etc. It is one of the most important virtual benchmarks used to validate the generalizability of virtual screening methods. | |||
'''Property-Unmatched Decoys in Docking Benchmarks''' -- [https://doi.org/10.1021/acs.jcim.0c00598 J Chem Inf Model (2021)] | |||
A more recent update of ligand+decoy benchmark, to cover property-matched and property-unmatched decoys. We use [https://en.wikipedia.org/wiki/Receiver_operating_characteristic semi-logarithmic Receiver Operating Characteristics curves] to see how ligands are favored over different decoy sets. | |||
'''A Database for Large-Scale Docking and Experimental Results''' -- [https://doi.org/10.1021/acs.jcim.5c00394 J Chem Inf Model (2025)] | |||
This is a real Large-Scale Docking benchmark with docking scores, as well as, experimentally-measured binding affinity, all in: https://lsd.docking.org | |||
LSD database hints to real binders and non-binders across 11 targets actively studied in the lab. | |||
== On the theory behind DOCK's algorithms == | |||
'''Automated docking with grid-based energy evaluation''' -- [https://doi.org/10.1002/jcc.540130412 J Comput Chem (1992)] | '''Automated docking with grid-based energy evaluation''' -- [https://doi.org/10.1002/jcc.540130412 J Comput Chem (1992)] | ||
How DOCK scores molecules. Rather than calculating every protein-ligand interaction from scratch, protein contribution to interaction energies (van der Waals, electrostatics, desolvation) are precomputed on a grid to be independent of ligands (by approximating), then when docking, the ligand atoms simply look up protein contribution to energies at the nearest grid points. This trick allows docking to be fast enough to screen at the scales we do. This paper does not include desolvation, as it was added later in. Speaking of! | How DOCK scores molecules. Rather than calculating every protein-ligand interaction from scratch, protein contribution to interaction energies (van der Waals, electrostatics, desolvation) are precomputed on a grid to be independent of ligands (by approximating), then when docking, the ligand atoms simply look up protein contribution to energies at the nearest grid points. This trick allows docking to be fast enough to screen at the scales we do. This paper does not include desolvation, as it was added later in. Speaking of! | ||
'''Rapid context-dependent ligand desolvation in molecular docking''' -- [https://doi.org/10.1021/ci100214a J Chem Inf Model (2010)] | '''Rapid context-dependent ligand desolvation in molecular docking''' -- [https://doi.org/10.1021/ci100214a J Chem Inf Model (2010)] | ||
Explains how DOCK accounts for energetic penalty of stripping water away from a ligand as it binds. Full water to octanol desolvation energetic penalty (so-called full transfer desolvation) of each atom is pre-computed at ligand building stage. Then at docking time, instead of applying a flat penalty, each ligand atom's desolvation cost is a fraction of the full transfer cost, scaled by how buried it is in the binding site. So, a polar group out in solvent is penalized less than one buried in a hydrophobic pocket. | Explains how DOCK accounts for energetic penalty of stripping water away from a ligand as it binds. Full water to octanol desolvation energetic penalty (so-called full transfer desolvation) of each atom is pre-computed at ligand building stage. Then at docking time, instead of applying a flat penalty, each ligand atom's desolvation cost is a fraction of the full transfer cost, scaled by how buried it is in the binding site. So, a polar group out in solvent is penalized less than one buried in a hydrophobic pocket. | ||
'''Ligand pose and orientational sampling in molecular docking''' -- [https://doi.org/10.1371/journal.pone.0075992 PLoS One (2013)] | '''Ligand pose and orientational sampling in molecular docking''' -- [https://doi.org/10.1371/journal.pone.0075992 PLoS One (2013)] | ||
How DOCK places our pre-built ligand conformers into protein pocket. It is a deterministic (non-random, repeatable) algorithm that given a set of user defined "matching spheres" that dot the protein surface, the algorithm finds possible ligand pocket placements by matching internal distances of the protein matching spheres (protein pocket curvature shape) to the internal distances in ligand's rigid fragment atoms (ligand shape). | How DOCK places our pre-built ligand conformers into protein pocket. It is a deterministic (non-random, repeatable) algorithm that given a set of user defined "matching spheres" that dot the protein surface, the algorithm finds possible ligand pocket placements by matching internal distances of the protein matching spheres (protein pocket curvature shape) to the internal distances in ligand's rigid fragment atoms (ligand shape). | ||
== On medicinal chemistry == | == On medicinal chemistry == | ||
| Line 38: | Line 64: | ||
This is an exhaustive guide towards reviewing the interaction of small molecules with proteins. Especially useful are the discussion of ideal bond lengths, angles, and energies. A must-read before (and during... and after... tbh just read this every few months) hit-picking from a virtual screen. | This is an exhaustive guide towards reviewing the interaction of small molecules with proteins. Especially useful are the discussion of ideal bond lengths, angles, and energies. A must-read before (and during... and after... tbh just read this every few months) hit-picking from a virtual screen. | ||
'''A Real-World Perspective on Molecular Design''' -- [https://doi.org/10.1021/acs.jmedchem.5b01875 J Med Chem (2016)] | '''A Real-World Perspective on Molecular Design''' -- [https://doi.org/10.1021/acs.jmedchem.5b01875 J Med Chem (2016)] | ||
A good read after "A medicinal chemist's guide to molecular interactions". The paper is full of real-world strategies, problems, and surprises in ligand optimization. All examples are based on prospective computational drug design. Especially helpful to read before analoging your hit series. | A good read after "A medicinal chemist's guide to molecular interactions". The paper is full of real-world strategies, problems, and surprises in ligand optimization. All examples are based on prospective computational drug design. Especially helpful to read before analoging your hit series. | ||
'''Development of a random background to understand ligand optimization''' -- [https://doi.org/10.1038/s41586-026-11013-5 Nature (2026)] | '''Development of a random background to understand ligand optimization''' -- [https://doi.org/10.1038/s41586-026-11013-5 Nature (2026)] | ||
| Line 47: | Line 75: | ||
What changes are most likely to improve the activity of a small molecule? And, how do we know whether a new method is better than random chance? This work sets out to establish a baseline for comparison, and provide insight into what modifications prove most useful for a given goal (increased potency, better PK, etc). | What changes are most likely to improve the activity of a small molecule? And, how do we know whether a new method is better than random chance? This work sets out to establish a baseline for comparison, and provide insight into what modifications prove most useful for a given goal (increased potency, better PK, etc). | ||
''' | '''Predicting new molecular targets for known drugs''' -- [https://doi.org/10.1038/nature08506 Nature (2009)] | ||
The Similarity Ensemble Approach (SEA) is a computational tool developed by Mike Keiser during his PhD in the lab, and is still used to predict potential targets/off-targets for compounds of interest. | |||
'''Seven Year Itch: Pan-Assay Interference Compounds (PAINS) in 2017─Utility and Limitations''' -- [https://doi.org/10.1021/acschembio.7b00903 ACS Chem Bio (2019)] | |||
Pan-Assay INterfering Substances (PAINS) are chemical moieties which recurrently appear across assays, but whose actions are determined by non-specific mechanisms. These mechanisms can include aggregation, autofluorescence, and reactivity with target proteins (among other mechanisms). There are shared structural motifs that can underlie these properties; our lab has worked extensively on characterizing compounds that aggregate, compounds with large conjugated systems can autofluoresce, and electron-withdrawing groups can incite electrophilic reactivity. Within the realm of virtual screening we find that there are other substructures (triazines, non-physiological protomers) which appear time and time again. In brief: Some chemical substructures suck, and we should avoid them. Baell's paper discusses some of the limitations of structure-based filtering, and when/how we should apply it. | |||
[[Category:Papers]] | [[Category:Papers]] | ||
Latest revision as of 20:25, 26 October 2026
The papers here are not exhaustive by any means, but are meant to serve as references for new members of the lab. Feel free to add additional papers here; alternatively, consider adding more specialized papers or applications of DOCK to Classic Dock References or Greatest Hits respectively.
On using DOCK for virtual screening
Virtual screening of chemical libraries -- Nature (2004)
Commentary written by BKS over two decades ago. Even to this day, we face the same challenges and celebrate the same successes outlined in this paper. The ideas covered here have become recurrent themes in our lab, and paint with broad strokes a research enterprise that has deviated remarkably little.
A practical guide to large-scale docking -- Nature Protocols (2021)
Discussion of general purpose concerns with regards to virtual screening, and guidelines for ensuring adequate control calculations. An incredibly useful resource. See also the Supplementary Information, which provides explanation for each row found in an INDOCK file (one of the core input files for DOCK 3.x).
A paper on ZINC, whether the original, ZINC16, ZINC20, ZINC22, or another -- tbd which one
The impact of library size and scale on virtual screening -- Nature Chem Bio (2025)
Bigger is better, and this paper demonstrates that screening larger libraries increases the likelihood of finding high-affinity hits. Just as importantly, this paper looks at how many compounds must be purchased/tested in order to reliably determine the hit rate from a given screen.
Read the dock docs -- Available here
The dock documentation contains the minimal instructions needed to run DOCK on our clusters, including commands that you can copy/paste in your own work. Functionally, our standard operating procedures for docking.
On benchmark sets
Directory of Useful Decoys, Enhanced (DUD-E): Better Ligands and Decoys for Better Benchmarking -- J Med Chem (2012) A rich set of ligands and decoys across 102 target proteins, spanning GPCRs, ion channels, kinases and etc. It is one of the most important virtual benchmarks used to validate the generalizability of virtual screening methods.
Property-Unmatched Decoys in Docking Benchmarks -- J Chem Inf Model (2021)
A more recent update of ligand+decoy benchmark, to cover property-matched and property-unmatched decoys. We use semi-logarithmic Receiver Operating Characteristics curves to see how ligands are favored over different decoy sets.
A Database for Large-Scale Docking and Experimental Results -- J Chem Inf Model (2025)
This is a real Large-Scale Docking benchmark with docking scores, as well as, experimentally-measured binding affinity, all in: https://lsd.docking.org
LSD database hints to real binders and non-binders across 11 targets actively studied in the lab.
On the theory behind DOCK's algorithms
Automated docking with grid-based energy evaluation -- J Comput Chem (1992)
How DOCK scores molecules. Rather than calculating every protein-ligand interaction from scratch, protein contribution to interaction energies (van der Waals, electrostatics, desolvation) are precomputed on a grid to be independent of ligands (by approximating), then when docking, the ligand atoms simply look up protein contribution to energies at the nearest grid points. This trick allows docking to be fast enough to screen at the scales we do. This paper does not include desolvation, as it was added later in. Speaking of!
Rapid context-dependent ligand desolvation in molecular docking -- J Chem Inf Model (2010)
Explains how DOCK accounts for energetic penalty of stripping water away from a ligand as it binds. Full water to octanol desolvation energetic penalty (so-called full transfer desolvation) of each atom is pre-computed at ligand building stage. Then at docking time, instead of applying a flat penalty, each ligand atom's desolvation cost is a fraction of the full transfer cost, scaled by how buried it is in the binding site. So, a polar group out in solvent is penalized less than one buried in a hydrophobic pocket.
Ligand pose and orientational sampling in molecular docking -- PLoS One (2013)
How DOCK places our pre-built ligand conformers into protein pocket. It is a deterministic (non-random, repeatable) algorithm that given a set of user defined "matching spheres" that dot the protein surface, the algorithm finds possible ligand pocket placements by matching internal distances of the protein matching spheres (protein pocket curvature shape) to the internal distances in ligand's rigid fragment atoms (ligand shape).
On medicinal chemistry
A medicinal chemist's guide to molecular interactions -- J Med Chem (2010)
This is an exhaustive guide towards reviewing the interaction of small molecules with proteins. Especially useful are the discussion of ideal bond lengths, angles, and energies. A must-read before (and during... and after... tbh just read this every few months) hit-picking from a virtual screen.
A Real-World Perspective on Molecular Design -- J Med Chem (2016)
A good read after "A medicinal chemist's guide to molecular interactions". The paper is full of real-world strategies, problems, and surprises in ligand optimization. All examples are based on prospective computational drug design. Especially helpful to read before analoging your hit series.
Development of a random background to understand ligand optimization -- Nature (2026)
What changes are most likely to improve the activity of a small molecule? And, how do we know whether a new method is better than random chance? This work sets out to establish a baseline for comparison, and provide insight into what modifications prove most useful for a given goal (increased potency, better PK, etc).
Predicting new molecular targets for known drugs -- Nature (2009)
The Similarity Ensemble Approach (SEA) is a computational tool developed by Mike Keiser during his PhD in the lab, and is still used to predict potential targets/off-targets for compounds of interest.
Seven Year Itch: Pan-Assay Interference Compounds (PAINS) in 2017─Utility and Limitations -- ACS Chem Bio (2019)
Pan-Assay INterfering Substances (PAINS) are chemical moieties which recurrently appear across assays, but whose actions are determined by non-specific mechanisms. These mechanisms can include aggregation, autofluorescence, and reactivity with target proteins (among other mechanisms). There are shared structural motifs that can underlie these properties; our lab has worked extensively on characterizing compounds that aggregate, compounds with large conjugated systems can autofluoresce, and electron-withdrawing groups can incite electrophilic reactivity. Within the realm of virtual screening we find that there are other substructures (triazines, non-physiological protomers) which appear time and time again. In brief: Some chemical substructures suck, and we should avoid them. Baell's paper discusses some of the limitations of structure-based filtering, and when/how we should apply it.