• Neuroscience & Pharmacology Discussion Welcome Guest
    Posting Rules Bluelight Rules
    Recent Journal Articles Chemistry Mega-Thread
    FREE Chemistry Databases! Self-Education Guide

  • N&PD Moderators: Skorpio

Evolutionary algorithmic program for molecular drugs

I'm on a VNC connection at the moment so I cannot watch this, but I will when I get a chance.

You have reminded me of a thought-experiment I've been picking ver for a long time.
First construct a program that makes a "model" of a molecular skeleton in memory. Provide it with a list of "attachment points" and possible substituents. For example, provide 2,5 dimethoxy phenethylamine, tell the program to attach e.g. halogens, methyl, ethyl, propyl groups on the 4-position of the ring and e.g. methyl, ethyl, hydrogen on the alpha carbon. Give it a run and it should do some valence checking and then output a combinatoric set of compounds for each possibility at each attachment point. Output can be as SMILES or some other structural descriptor.
Combinatorically generated compounds are energy-minimized and fed into a docking simulation en masse. Hopefully this would result in a big chart of e.g. predicted affinity vs. lipophilicity/charge/size that could be used to drive an evolutionary algorithm.


Edit: Holy shit this is basically the s/w in the video, or a similar concept (GUI drug designer). I dunno what it is though - Medical s/w is usually expensive as shit.

Time to hit the Java books I guess.
 
Last edited:
Holy shit. This is amazing; the ability to iterate through so many possibilities automatically can provide a far wider range of ligands than would otherwise be possible. This will be a quantum leap in combinatorial chemistry.

Just imagine hundreds of years (maybe less?) down the road when accurate structures of every human gene product are available and new drug structures can be checked against each and every one to determine its selectivity and effect profile.
 
It seems to me that our ability to design molecules based on pharmacological activity is limited by our understanding of their mechanisms of action. While a program such as this could definitely lead to exponential growth within the realm of chemical design, it would be very difficult to determine a compound's full range of effects (both physical and mental) from a computer program.
 
In all honesty I don't understand what this is supposed to be... the software looks cool but I think it may just be a trick video; they aren't really doing anything.
 
1. Start with a nucleus of a molecyle (fluoroquinolines, it looks like in the video)
2. Select attachment points
3. Select acceptable 'ranges' for molecular characteristics like logP, hydrogen bond donors/acceptors, pKa etc.
4. Computer 'attaches' functional group to the attachment point
5. compounds are scored according to the criteria and user input, start again at step 2.

Eventually you should have a compound that acts as a ligand with whatever properties you want to imbue it. In theory, anyway.
 
At the end of the video it states that the software used is "CoalesiX", by the featured company.
 
nice link, watched a bit of the video. the human-guided GA scoring is interesting, a lot of programs of this type use scoring functions with less human input. human input at each generation would be more time consuming, but might converge to useful (ie. synth-able) solutions. human intuition can be a double edged sword: a good medicinal chemist has rule-of-thumb about what functional groups are most useful based on previous drug discoveries (of which there are many). however, this same 'intuition' can introduce a bias that may lead to good candidate structures being overlooked because they don't 'look' like previous blockbusters. and did you see the way that human operator was blinkin in the beginning of the video? too many hours in front of monitor IMHO. but i'm sure none of us know anything about that. ; )

also, this reminds me of another drug design method where, given a known target structure (enzyme, receptor, etc.), with known pharmacophore (=features of a molecule necessary for a given action at given target), small molecule fragments (such as phenyl rings, alkyl groups, small polar groups, etc.) are docked tightly (but independently)to the places required by the target for activity. then, when those functional groups are in place, the program looks for a molecular "scaffolding" that will position those groups right where they need to be. kind of like, connect-the-dots, but connect-the-fragments. this has been successful in designing ligands with VERY high binding affinities and activities. although stronger is not always better, in big pharma it is valued. also there is merit in just coming up with a new scaffolding, in that often the new molecules will not have the same basic skeleton as known actives. this is called "scaffold hopping", leads to whole new series of analogs. so like if you wanted to do a serotonin 2a agonist for example, the program would probably dock a phenyl ring, a basic nitrogen, and maybe an oxygen or two to the receptor and then try to combine those fragments into a whole molecule, which would probably NOT be a phenethylamine, tryptamine, or ergoline. sometimes the structures it comes up with are impracticle to synth, but i think there are ways to screen those out.

oh well, happy researching.
nirvus
 
Originally posted by nirvus

also, this reminds me of another drug design method where, given a known target structure (enzyme, receptor, etc.), with known pharmacophore (=features of a molecule necessary for a given action at given target), small molecule fragments (such as phenyl rings, alkyl groups, small polar groups, etc.) are docked tightly (but independently)to the places required by the target for activity. then, when those functional groups are in place, the program looks for a molecular "scaffolding" that will position those groups right where they need to be. kind of like, connect-the-dots, but connect-the-fragments. this has been successful in designing ligands with VERY high binding affinities and activities. although stronger is not always better, in big pharma it is valued. also there is merit in just coming up with a new scaffolding, in that often the new molecules will not have the same basic skeleton as known actives. this is called "scaffold hopping", leads to whole new series of analogs. so like if you wanted to do a serotonin 2a agonist for example, the program would probably dock a phenyl ring, a basic nitrogen, and maybe an oxygen or two to the receptor and then try to combine those fragments into a whole molecule, which would probably NOT be a phenethylamine, tryptamine, or ergoline. sometimes the structures it comes up with are impracticle to synth, but i think there are ways to screen those out.
Click to expand...

Do you have any examples of drugs that have been designed by this method? Not trying to be a ball buster, I'm just curious.
 
apsig said:
Just imagine hundreds of years (maybe less?)
Click to expand...

Likely much less from the statistics I've been looking into. Twenty years? The exponential growth of information technologies and their application is astounding, and speeding up rather than slowing down.

MattPsy said:
At the end of the video it states that the software used is "CoalesiX", by the featured company.
Click to expand...

That's what I get for not reading credits. Thank you.
 
sekio said:
1. Start with a nucleus of a molecyle (fluoroquinolines, it looks like in the video)
2. Select attachment points
3. Select acceptable 'ranges' for molecular characteristics like logP, hydrogen bond donors/acceptors, pKa etc.
4. Computer 'attaches' functional group to the attachment point
5. compounds are scored according to the criteria and user input, start again at step 2.

Eventually you should have a compound that acts as a ligand with whatever properties you want to imbue it. In theory, anyway.
Click to expand...


Maybe this means I should watch the whole video before my dumb ass goes posting in the thread about it. :P
Click to expand...

Well if their video wasn't so poorly explained I could make it to the end without losing interest in the narrator's jargon about evolutionary principles and not the fact that they were promoting software. But even if I didn't watch the last 20%, using this program is fruitless to a knowledgeable molecule designer.

The requirements you list for their program are to know a binding site and binding contacts, which implies you have a crystal or NMR structure (or model from one) with docked ligand in hand. Any intelligent person trained in molecular design could easily accomplish this same task using their own brain and free structure viewing software, and have no need to buy, use or learn the Coalesix software to accomplish this.

Additionally a brain can incorporate other constraints such as ease of synthesis in design, as mentioned. It only really makes sense to use if you don't know what you're doing which means you probably shouldn't be doing this kind of work anyways.

I wonder what their price tag is...
 
It's big enough that it's not shown on their website.

The point is, software like this makes it much faster for people and computers to do the modelling and drug-design process. People could (and did) do Calculus all by hand too but now we have fancy computers to do so.
 
yeah sekio, we have computers to do calculus, and i'm glad we do, but that's not the same as being able to it by hand. same thing in a lot of subjects. technology makes it accessible but does it make us better at them? i can't talk, i've switched from chemistry to computational chemistry! push of a button.

said program prolly as some nice bells and whistles. and the business folks who run pharm research firms tend to like technology which automates more of the process of drug discovery, allowing them to hire an outside medicinal chem consultant for a few hours before filing NDA's (new drug application) vs. paying a post-doc to tinker for months on salary.

i think any chemistry software worth it's salt can define queries where you can search chemical data banks based on part of molecule known, where groups can be added, etc. I'm pretty familiar with Spartan and it's not a pharm program but it now does that as well as pharmacophore screening, etc. Well whatever, if you can afford it, kudos. Open source is coming on stronger all the time now, wonder when we'll see stuff like this. Ha, probably never. for obvious reasons. ..
 
You must log in or register to reply here.
Top