Medicine making puzzles

Started by Elfi

Elfi Lv 1

This has probably been supposed before. I’m rather new to the ligand puzzles, so I may have missed it.

I would really love if there was an option for the Ligand puzzles to have recipes that would do things like:

  • Try out one hydrocarbon at all possible spots - one at a time and take it out afterwards - then rank them after which placement scored highest. (Or remove one hydrocarbon)
  • Swap around 1 N with 1 O at all possible spots - one at a time and take it out afterwards - then rank them after the highest score. Another favorite is doing the exact opposite, starting with changing O’s to N’s.
  • Swap atoms not N, with N at all possible spots - one at a time and take it out afterwards - then rank them after the highest score. Could be used for other atoms too.
  • Add one hydrocarbon and then remove another. Try all options and then rank them after the highest score.
  • Increase ringsize by one (or decrease ringsize by one.)

One thing I have noticed is that it often pays to stick a hydrocarbon after an N if there is space. Same thing after an O.

I assume most of this is known already.

I try out several eg hydrocarbon positions while looking for the changes that make the least score drop. I will try them out one by one. When I have made a structural change, I do a wiggle and allow it to run until it seems to have reached its max. If the score is higher than my starter score, I accept it and go back to running recipes on the ligand again. If it is not higher, but still promising close to my original score, I look for other changes that may be worth making. Because regularly one can stack multiple changes together that on their own are not good enough to get a higher score, but together they are. Those can be any of the different kind of different structural changes. I could end with changes of remove one hydrocarbon, add one hydrocarbon, remove one N. When I get above my starter score, I go back to running regular recipes on the ligand.

Some of the atoms seem rarer in use. It would be particularly helpful to have a recipe that could try out these in a systematic way. But then also check that they don’t cause bad groups to happen.

Thx to ZeroLeak who’s molecule showed me that it is a good idea to stick hydrocarbons in different places where there is space to fill. I hadn’t tried that myself. Thx to Carxo who’s molecule was very different from mine, but better. It showed me that rings can be used for filling up space in many different ways. Also it made me realize that 4 ring size could be worth trying.

Thx for all the fine recipes that you guys have made.

Bruno Kestemont Lv 1

Support for this.
We spend too much time to try many atoms, rings etc. The result is that we don't try everything (also, not beeing scientists, we don't always have the sense of what can be relevant). Personally, I rarely try rare atoms because the chance to gain points is too small. There are potential missing solutions here.
Did doesn't diminish the role of our basic intuition or observation in desinging.

Bletchley Park Lv 1

I commend @Elfi for suggesting this and would recommend making this interface in a way that it operates using SMILES strings as input, this way we can modify and extend SMILES strings as input to the model at scale, by running this simultaneously on multiple clients with defined start and endpoints in the permuations on each client. Each client can then run the usual wiggle and shakes and other processing on each processed SMILES input string to see if that yields a gain. Top scoring SMILES strings can then be output (or saved in quicksaves) for further processing.

Bruno Kestemont Lv 1

Indeed could the folowing proposed commands could do the job ?

string structure.GetSmile(integer segmentIndex)
void structure.SetSmile(integer segmentIndex, string SMILE)

Elfi Lv 1

Thx for the support, @"Bruno Kestemont" and @"Bletchley Park"!

I now have a few additions to what I do above. The first bit could be related to the automation I was wishing for. In that it could be trying out options for ring making. The second part would only be automation related, if one could specify blocks of recipes to run after each other. Like blocks of mini programs, where a block would be a recipe. (Eg. I sometimes like to run cycles of BlueFuze 2020 and BlueFuze 2020b and they are fairly short runned) Anyway, I mention this as a guide to what can be done when working on a ligand, while there are multiple other ways.

MAKING RINGS

I have realized that when I have enough hydrocarbons (methyl groups) close together (two or more), sometimes it pays adding one or two additional hydrocarbons and connecting them to create a ring. It is a kind of taking advantage of what direction the molecule itself decides to grow in.

RUNNING RECIPES

There are also specific recipes I have more success with using on ligand puzzles than others. In the beginning I just threw every recipe after the ligand because I didn't know what worked. Sometimes I could get a recipe to run and other times not. I think it may depend on if the markers were on the ligand or the protein. I had other recipes working as well. But these are the main recipes I used to up the score.

When I have made a structural change I typically run Acid Tweeker V2.8.1. https://fold.it/recipes/101634

After a while I check if I can get new structural changes in. If not I run it again.

On other recipes that work for me on ligand puzzles, I have had particular luck with these two:

These two are particularly helpful in cases where there is a very loose fit between the ligand and the protein. Like after extensive structural changes where I have made many changes in one go. These recipes seem to help the ligand grab hold on some of the side chains that can form hydrogen networks. I also use them later on if I can't get anywhere with Acid Tweaker or modifying. Sometimes they manage to squeeze the ligand and protein to a better fit so I find somewhere I can modify. I turn on IsoSurface to see where there is space left.

When I have run Ligand Docker 2.5d and gained a raise, I immediately stop it. Because it doesn’t seem to gain further score raises afterwards. Then I sometimes start it again. If I see the solutions moving around a lot, I think this is suggesting there is more space for a better fit or more stable structure. So I either run more recipes or try to put in additional structure.

Just as important as fitting more stuff into the ligand, it is to watch what bumps against the IsoSurface and take out structures that get too close. This may change after a structure addition and recipe run as the ligand gets rearranged a bit in space.

Similarly with AFK3.5.1 When the amount of points it makes gets negligible, I stop it. When these two recipes don’t help anymore and I still can’t make structure changes, I return to Acid Tweaker.

If I still can’t get structure score changes above what I have, I try to make structure changes that get close to the score I have and then use a recipe to get them past. Sometimes this will open new options. However I mainly employ this strategy late and as a last resort.

Also sometimes when there are two spots on the same Carbon for placing things, it sometimes matters which order they are placed in. I have noticed it particularly when I place a hydrocarbon at one spot and an oxygen at another. I tried the different options and I didn’t always get the same score depending on which I placed first.

Floddi Lv 1

@"Bruno Kestemont" Nearly every "View Protein" option isn't a great way to understand it but "Sphere". Unfortunatly you cannot see much when using it but it depicts the real atom radii. Halogens (F<Cl<Br<I, see PSE) are sometimes better than hydrogens because they fill out more space. But they tend to be partially negative and it can interact with groups being part of the same aromatic ring (I don't know if FoldIt consideres this scorewise). They can be used in areas were a methylgroup would be suitable. Triflates (R-CF3) can be used as a more lipophilic group with a bigger surface area than methyl groups. They should be used in para-position (benzene)

S can be good in ring systems of 5 and makes the structure polar but not as much as smaller molecules with a lower surface area.
The "Fragment Selection" has everything which can be used. Sometimes a hetero cycle is too big or the distribution of charges. Sulfoxides (I think 2709 also has it) can be useful if you have two positivly charged groups facing each other.

I've never used P but as Phosphate it should be stable. Maybe it will also work with the Arg's in 2709 :)

Bruno Kestemont Lv 1

I just published my 3 favourite "fit for ligand" recipes:
1) Acid Tweeker V2.8.2.
It's very effective at end game. Default also moves the ligand, starting with the ligand. To run overnight.

2) A "light" version of Ligand Docker, Ligand Docker v2.5.1 light that I use preferably to its parent. It is designed to work overnight. The dialog is smaller, and when stopping, it sorts in the "undo's" and in the log, all attempts made with score, bonus and number HBonds. It's very usefull to "manually" select the best one before to go on (but it takes quite a long time to stop if there were many solutions).

3) A specific GAB for Ligand puzzles that I also use overnight when I have a good start (it's necessary to call credited best afterwards because the recipe doens't stop at the higher score). It systematically moves random atoms of the ligand, which can gain many points when it finds a better torsion or surface. Most gains are in the first rounds, but other big gains can also occur later on.