Dear all,
The field of CADD has become young. Now it is required that there should be an open source CADD platform. This platform should contain the prepared protein for drug targets and proteins responsible for toxicity of drug molecules. It should also have all docking methods available. Therefore, for each query results can be obtained for affinity towards all protein and all targets. The toxicity data can also be assessed in one platform.
This will make drug discovery faster and cheaper. What is your view?
ya it is very right to say that if possible then it will turn the fate of drug discovery to a major extent.
Idea seems great and one day it will be implemented. If such platform is made available the drug discovery process will see its fine days. Though lot of hurdles are there development of such platform would become another area...
I thing that something similar to what you are envisaging is already taking place. Or at least it is going in that direction.
Still something I need to investigate further but have a look at the OPEN PHACTS discovery platform developed by a consortium of Pharma and Academic groups.
From their web site:
"To reduce the barriers to drug discovery in industry, academia and for small businesses, the Open PHACTS consortium is building the Open PHACTS Discovery Platform. This will be freely available, integrating pharmacological data from a variety of information resources and providing tools and services to question this integrated data to support pharmacological research."
In an ideal world, it seems like a pretty good idea.
In real life, next to the questions it poses with respect to IP and data safety, I see a number of pitfalls in this approach. I could discuss for (very) long, but I'll try to be short.
First, there's the multiplicity of the tasks... "the prepared protein for drug targets and proteins responsible for toxicity of drug molecules" means a lot of potential receptors, and you'd multiply that by "all docking methods available". That already means an awful lot of information to host, and of calculations to run.
Still, this isn't enough... not all relevant targets/proteins will have a structure available. What about those? On the other hand, several targets are quite flexible and can have different conformations (examples: open/closed conformations of ion channels, DFG in/out conformations in kinases,...). If you start multiplying the number of drug targets by the number of (potentially) relevant conformations, it adds a further layer of complexity.
Next, there is the validation aspect. The usual situation for computer-aided drug design is "Crap in, crap out". Usually, most docking methods or scoring functions perform well for some types of compounds / targets, but not for others. All models need validation (or, at least, I wouldn't trust results from models/methods that were not validated). All training sets need to be cleaned. That is again an awful lot of work.
Another question is the degree of control the users will have about parameters. What defines a hit? How accurate should the docking parameters be? Do we dock over the whole protein, or only in a specific binding site? Should some protein flexibility be taken into account? There is probably a lot of things to consider here too.
Finally, regarding affinity: score functions in docking programs are optimized towards prediction of the binding mode. They are (more or less) OK for that, but they often fail to estimate compound activity. This can already be a problem when handling a single target. When one starts to compare activities that are suggested by such functions over multiple targets, how reliable will this be?
I'll stop at this point, and share two recent experiences I had...
The first one is testing an online docking service (so, this is similar to what you propose). They had two models of Rho Kinase available there, from crystal structures. I picked up the one that was complete (part of a loop was broken in the other one) and submitted one of the molecules from a recent paper we wrote with my co-workers. I fixed the correct protonation state myself.... Still none of the proposed poses were correct, even though the model was OK (from a crystal structure) and the docking program was OK (actually a more evolved version of what I use). So, I assume the problem is in either docking parameters or in the particular conformation of the protein in that structure.
The other experience was reviewing a selection of compounds proposed by a commercial compound supplier, based on docking. They used Glide for docking, which has good reputation. And there were plenty of structures available for that target. Still, I would not even remotely consider testing 98% of the proposed compounds. Again, the model and parameters probably need some work / validation.
So, this is why I'm not fully enthusiastic about the idea of an automated, open source CADD platform.
Dear Sandro,
You are absolutely right.............
but I think we should consider following points............
1. Besides all failures, CADD is providing drug molecules.
2. Are you personally not using CADD tools.
3. We should be solving problems rather than stopping?
4. I know these problems will arise but together we can solve them. Because every problem has a solution.
5. We are not saying to completely rely on the CADD solutions. But the consensus results are more accurate than individual results. If we obtain similar poses with most of the docking methods we can at least rely on those poses.
6. I don't think that we consider protein flexibility for each and every target.
7. We cannot have all the software to be included but we can club at least freeware and those who are willing to add add some feature to this program.
So the hurdles are there but a collective effort can clear them........
I agree with you it's good idea and this kind of platform will help those who are interested in the field of drug discovery..
- Badges
- Science topic
- Similar topics
- Thesis Research