Devprev Drug Design Update
I want to thank you for taking the time to test the first drug design puzzle. It is very nice to be part of a community that is so enthusiastic and helpful in tracking down bugs, reporting them, and improving science. I know that it can be extremely frustrating testing things that are broken or that do not perform in logical ways. Thank you for sticking it through and reporting the bugs and feature requests.
The idea behind releasing the puzzle to devprev without any information was to get a handle on what needed to be done to improve the drug design module without the constraint of a specific problem being solved. In the future, this will not be the case and a detailed description of the puzzle and the disease you are working on will be given.
This puzzle only featured one small aspect of the drug design tools and modes. In this puzzle, you were allowed to only change the atom identity to a Carbon or Nitrogen. By limiting the tools, we were able to identify specific problems with drug design mode (see below for summary). This is not what the future is of drug design. Adding large and small fragments, extending the small molecule by specific atoms, designing through synthetic rules, and queuing ligands are all tools that were not available. To let everyone get a peek at what is to come, jflat has created another sub group that has all the tools available. We are currently testing this build, but look for a blog post in the future where we will give you instructions on how to obtain it.
Here is a summary of the bugs/features that were reported.
We have tutorials! We just have not released the tutorials. The tutorials take you through a scientific paper where rational drug design was used to create an inhibitor for the FK506 binding protein. The tutorial puzzles will be in the new experimental drug design group that is being set up.
There were a lot of bugs/features about the ligand grids. A lot of people wanted the color of the grids to be changed. Is it better to have a simple color change is needed or the ability to change the color on the fly is needed?
The ligand grids were put in place to show contact points where you can design acceptors/donors atoms or fragments. Right now, the grids only show acceptors, donors, or repulsive scores. The future of the grids is to include information about structure activity relationships (SAR) of small molecules. SARs help identify regions of the small molecule or protein that like specific features, like hydrophobic/polar atoms, bulky/small fragments, etc. SARs are unique to each protein, so the ligand grids will change.
Finally, there were numerous bugs with the grids. The grids would disappear/appear randomly, or move, or not show when toggled and the slider was moved. These bugs will be fixed!
Ligand jumping out of pocket
I believe that I have tracked down the problem for this. Interestingly, I believe it is how I set up the ligand center filter, which was supposed to keep the ligand in the binding pocket. You have to love irony.
Where to begin? The idea for limiting wiggle, which is what the puzzle was supposed to do, is that drug design is usually done with a fixed backbone with no wiggles/shakes. While this is the general rule of thumb, there are optimization steps done to the backbone and sidechains throughout the design process. The problem with this specific puzzle, and why we wanted to limit wiggle/shake, is that the binding pocket was extracted from a larger protein. There were cutpoints at the end of each helix. Through internal testing, the protein would explode each time I wiggled the protein. I thought that I had fixed the issue, but it looks like there were some problems left.
For the future, the current plan is we will not limit wiggle or shaking. One of the goals of the Foldit community is to develop new strategies to fold proteins. For drug design, we want to keep in alignment with those goals and allow free thought and movement of the protein. New puzzles will not have these constraints associated with them (unless they are really needed).
One thing not shown in this puzzle is on the fly rotamer development. This means, each time you modify the ligand, a whole set of new rotamers are generated. This feature is enabled in the new experimental group.
Scoring is notoriously difficult for small molecule drug design. For this puzzle, we did not make any modifications to the score function to make it easy to distinguish which small molecule was the best. There is a new filter that we have developed that better recapitulates experimental values, but it was not included in the puzzle. Speaking of filters, there will be a whole lot of them introduced in the future!
We look forward to your additional feedback!( Posted by free_radical 79 1868 | Fri, 12/04/2015 - 17:49 | 0 comments )
Let's Get Ready for Drug Design Puzzles!
The first drug design puzzle is on its way! This will likely take place next week (unless of course, we run into currently unforeseen issues**).
I would like to first tell you why I am particularly excited about this series of puzzles. During the past decade, advances in drug design (specifically structure-based drug discovery) have seen a lot of success. However, the automated tools for this are encountering a huge problem with the total chemical space that can be sampled. It is estimated that the theoretical possible drug-like molecules is 1030-1060. This means that more than 99.9% of all drug-like molecules remain to be synthesized and explored for their therapeutic benefit. The scale is huge!
This is where you, as Foldit players, can help. The immense chemical and conformational space that needs to be sampled is greatly reduced with human spatial recognizing skills, which are far more robust at recognizing patterns than computer algorithms and can intuitively sample conformational and chemical space.
There is a lot of new technology involved with the drug design puzzles – it ranges from new drug design algorithms to robots! We will start covering these topics as the puzzles are introduced. For now, we are introducing a series of simple puzzles to help you understand some of the new tools. As time passes, we will release more complicated drug discovery problems and will start to provide targets with the hope of fighting a specific disease. The diseases that we will work on are rare and neglected diseases; these diseases have little research done on them and represent a space where the Foldit community can make a huge impact. If you have time, check this article out on the growing problem of rare and neglected diseases.
We look forward to hearing your feedback on these puzzles!
** which we did but we're ironing out!( Posted by free_radical 79 1868 | Tue, 09/29/2015 - 15:49 | 4 comments )
It’s almost time for class to be in session once more, and we want to welcome all our new students (academic and from the school of life) to Foldit! We have put together some handy tips to make your playtime in Foldit enjoyable.
Pick a good name but remember it is visible to everyone! Remember it should fit within the community guidelines, but more importantly, should reflect you - but not to the point where you’re using your email address as a login ID. That is just a bad idea for internet safety in general, so please don’t do this. Keep your name PG and something you’d be proud to see on top the leaderboards. If it’s too late and you think you've chosen badly with your username already - I can help you fix that. No judgement from me, we all veer off into “thought it was funny at the time” land on occasion.
Be patient! Foldit has an amazingly steep learning curve. A lot of our players are helpful, but being a global community, they may not always be watching the chat channel to answer questions. Have you checked our wiki? Did you try resetting, rereading the tutorial and trying again? Were you aware of the vast range of videos available you could watch? These are all great first steps but if they don’t get you what you need to know, then ask!
If you are having technical issues with a puzzle, please post them in the puzzle thread that goes with the puzzle you're having trouble with! This is the best way for our scientists and developers to find your issue. If you want to join in on some great learning threads, head to the forums!
We’re so glad you could join us and hope to see you around the community soon.( Posted by inkycatz 79 1868 | Fri, 09/04/2015 - 17:44 | 0 comments )
Drug design puzzles coming your way!
One of the problems that drug designers face seems innocuous at first: what would happen if a single atom in the small molecule changed from a carbon to a nitrogen or vice versa? Will this make the drug bind better to my protein target? At first glance, if you just look at the properties of the elements it might seem like an easy solution. Carbon in small molecules is mostly inert and “hydrophobic” while nitrogen is a “polar” atom and can either accept or donate hydrogen bonds. If a crystal structure is available, drug designers will replace a carbon with a nitrogen to form a hydrogen bond, which should increase activity. However, this can cause a conformational change in the ligand and reduce binding or the chemical properties change in such a way that it is hard to predict.
As you start to embark on your on drug designs for protein targets, you will be faced with a similar challenge. To this end, we have created a dataset of crystal structures with small molecules bound that have a single atom change, a carbon to a nitrogen or a nitrogen to a carbon. As your first drug design puzzle, you will be tasked to determine at which positions to change in order to increase binding to the target. To add on to the excitement, we have given the exact same dataset to our drug designers. At the end of the series of puzzles, we will post the results to see who did better!
Since these will be the initial drug design puzzles, we are looking for a lot of feedback. Functionality is limited in these puzzles (don’t worry, we will give you many more tools in the future and new gameplay modes!) in order to do a thorough testing of the interface. If we can, we will implement the requests that you make.
What is coming in the future?
New tutorials: Follow the steps of a group of scientist as they developed a new small molecule
New gameplay modes: Free design and synthesis based design
New media: chats with scientists and thoughts on the future of drug design
Questions? Add them here and we'll keep them in mind to answer in future posts!( Posted by free_radical 79 1868 | Wed, 07/22/2015 - 15:39 | 2 comments )
Feedback on Marburg puzzle 1108 ("Compact 37-Residue Marburg Virus Inhibitor Design")
Hi, folks. Thanks to all who played the latest Marburg puzzle. We have looked over the results, and it looks like the scoring was promoting some of what we wanted to see, though I think that maybe we should have given a bigger bonus for creating a core (and maybe required a few more residues in the core). While the top-scoring designs looked pretty good, some of the most interesting were, once again, in the "shared with scientist" category. Two in particular stood out:
Susume of Anthropic Dreams created a very interesting-looking sandwich of beta-strands, with a great "leapfrog" arrangement of disulfide bonds (cys3-cys21, cys7-cys33). Importantly, all strands contributed hydrophobic residues to the core, and there were no voids in the core, so it's a plausible-looking fold for presenting the antibody loop. My one criticism is that this only makes one additional charge-charge interaction with the target and one additional hydrophobic interaction, but on the other hand, putting the effort into having a nice, stable fold might be a good strategy.
LociOiling from the Beta Folders went with a nice, classic helix packing against a three-stranded sheet. The helix is very nice and plausible, with one clearly hydrophobic face in which every turn either presents a hydrophobic amino acid residue, or is involved in a disulfide bond. Again, we have a nice "leapfrog" disulfide pattern of the sort often seen in small peptides in the natural world. In this case, my only criticism is that the third strand isn't doing much to contribute to the fold or to binding (with the exception of a single valine that's making a hydrophobic interaction with the target). Still, it's conceivable that this would bind just fine if it were truncated down to a helix and a two-stranded sheet...
...Which is a good segue into our next puzzle: a smaller, 25-residue Marburg binder. Why smaller? Stay tuned for the new puzzle for more information!