Chemical Structure Generation for Drug Simulation
47 views
Skip to first unread message
Gokula Krishna
Mar 13, 2015, 1:43:56 PM3/13/15
to diy...@googlegroups.com
Hello there,
I am a python programmer who is interested in automating the drug simulation process. I taught myself about the basics of:
1. Molecular Dynamics
2. Pharmaco Kinetics
3. Pharmaco Dynamics
4. PyMOL
I am planning to create a program that generates all possible combinations of chemicals for the small molecule drug for a given protein. I could not find any source related to the following:
1. How to generate all possible combination of chemicals of drug for the protein? And how the generated chemicals should be represented for the protein-chemical Interaction (chemical structure like CH4 or other format)?
2. How to determine whether the given chemical is an agonist or antagonist computationally? And how to determine whether the viral protein is deactivated for the particular chemical?
3. An example on interaction between a simple viral protein and a chemical drug which deactivates it.
Thanks,
Tom Weingarten
Mar 14, 2015, 11:52:32 AM3/14/15
to diy...@googlegroups.com
Hi,
First off, a word of caution - the problem you're looking at here is worthy of multiple PhD theses. That being said, ambition is good, right? :)
1. When you say you want to generate all possible combinations of chemicals, you'll need to be more specific with your question. Do you have a specific small molecule drug compound in mind, and you want to generate different conformations of that compound? Or do you want to traverse through the space of all possible small molecule drugs that could exist? If the latter, I would suggest that this is impossible, because the space approaches infinity combinatorically rather quickly as you increase the number of allowed atoms.
A good algorithm that searched this space intelligently, however, and could provide suggestions for new, unknown small molecule chemicals, could be quite beneficial.
2. In order to predict activity with MD, you need to have prior experimental knowledge of what protein states are active and which are inactive (at least as far as any studies I've seen). This is because the activity of the protein is not something that is immediately obvious outside of its context. This is especially true of proteins that interact with other proteins or have activity that occurs over the time scale of milliseconds or longer.
Instead, researchers usually target a known binding site, and try to measure binding affinity for their target compounds. This is a costly MD experiment to run, so usually it is only done on a relatively small number of compounds, following a heuristic funnel that reduces the number of targets from millions to dozens.
3. For examples of proteins and their interactions - have you tried the PDB? http://www.rcsb.org/pdb/home/home.do
Best,
Tom
PhD, Computational Biology, NYU '12
First off, a word of caution - the problem you're looking at here is worthy of multiple PhD theses. That being said, ambition is good, right? :)
1. When you say you want to generate all possible combinations of chemicals, you'll need to be more specific with your question. Do you have a specific small molecule drug compound in mind, and you want to generate different conformations of that compound? Or do you want to traverse through the space of all possible small molecule drugs that could exist? If the latter, I would suggest that this is impossible, because the space approaches infinity combinatorically rather quickly as you increase the number of allowed atoms.
A good algorithm that searched this space intelligently, however, and could provide suggestions for new, unknown small molecule chemicals, could be quite beneficial.
2. In order to predict activity with MD, you need to have prior experimental knowledge of what protein states are active and which are inactive (at least as far as any studies I've seen). This is because the activity of the protein is not something that is immediately obvious outside of its context. This is especially true of proteins that interact with other proteins or have activity that occurs over the time scale of milliseconds or longer.
Instead, researchers usually target a known binding site, and try to measure binding affinity for their target compounds. This is a costly MD experiment to run, so usually it is only done on a relatively small number of compounds, following a heuristic funnel that reduces the number of targets from millions to dozens.
3. For examples of proteins and their interactions - have you tried the PDB? http://www.rcsb.org/pdb/home/home.do
Best,
Tom
PhD, Computational Biology, NYU '12
Supreet Kaur
Mar 18, 2015, 7:19:47 AM3/18/15
to diy...@googlegroups.com
certainly it sounds like a mammoth project but it is definitely possible and need of the hour to develop search automation models
to increase the pace in new age drug research. rather than measuring the project the scale of no of phd thesis I suggest this could be
quickly done by aggregating efforts of passionate people from both it and biotechnology as a collaborative sourcing (opensource )
what if you thoughts i am also putting my intent to participate in such opensource project..
coming to me i did my masters in biochemistry and i have working knowledge in python
good luck
Supreet Kaur
Gokula Krishna
Mar 30, 2015, 10:39:25 PM3/30/15
to diy...@googlegroups.com
I have a Molecule in SMILES format, eg., Zidovudine: Cc1cn(c(=O)[nH]c1=O)[C@H]2C[C@@H]([C@H](O2)CO)N=[N+]=[N-]
And the PDB file for HIV. How to calculate binding affinity computationally? Is there any API?
On Friday, March 13, 2015 at 11:13:56 PM UTC+5:30, Gokula Krishna wrote:
Tom Weingarten
Mar 31, 2015, 10:08:12 AM3/31/15
to diy...@googlegroups.com
AFAIK there's no API for that. If there were, I suspect it would be forced to make a lot of assumptions that would sacrifice a significant deal of accuracy for speed and ease-of-use. Typically this type of experiment requires a good deal of hand-holding, as there are decisions that need to be made wisely at many steps of the project.
For instance, even once you have the PDB file for HIV Protease, you need to make decisions about how to represent Hydrogens on some of the more volatile amino acids, based on such considerations as PH.
Also, you mention having the PDB file for HIV. Do you mean to say a specific HIV protein? I guessed protease because that's very common. Simulating an entire HIV capsid is a massive task (see: http://www.isgtw.org/feature/64-million-atom-simulation-%E2%80%93-new-weapon-against-hiv). Additionally, performing a binding affinity experiment in the context of an entire capsid is unnecessary - generally these simulations only include the protein and small molecule in question. Although it's possible that interactions from other proteins may affect the binding affinity, performing this sort of simulation across an entire capsid would be impossible, because you'd need an extremely long simulation at a massive scale to account for the flexibility and size of the system.
Best,
Tom Weingarten
For instance, even once you have the PDB file for HIV Protease, you need to make decisions about how to represent Hydrogens on some of the more volatile amino acids, based on such considerations as PH.
Also, you mention having the PDB file for HIV. Do you mean to say a specific HIV protein? I guessed protease because that's very common. Simulating an entire HIV capsid is a massive task (see: http://www.isgtw.org/feature/64-million-atom-simulation-%E2%80%93-new-weapon-against-hiv). Additionally, performing a binding affinity experiment in the context of an entire capsid is unnecessary - generally these simulations only include the protein and small molecule in question. Although it's possible that interactions from other proteins may affect the binding affinity, performing this sort of simulation across an entire capsid would be impossible, because you'd need an extremely long simulation at a massive scale to account for the flexibility and size of the system.
Best,
Tom Weingarten
Reply all
Reply to author
Forward
0 new messages