Side effects on every side

 

Skolnick, along with collaborator Mu Gao, produced a set ofartificial proteins through computer simulations, which were folded accordingto the laws of physics rather than evolutionary optimization. They compared thepairs of binding pockets and statistical significance of their structuraloverlap in both the generated proteins and natural proteins, and found that theartificial pockets had corresponding pockets on natural proteins. The modelingwas done with a representative covering of all protein types. 

 

"You could have the same or very similar pockets on the sameprotein, the same pockets on similar proteins and the same pockets oncompletely dissimilar proteins that have no evolutionary relationship. Inproteins that are related evolutionarily or that have similar structures, youcould have very dissimilar pockets," Skolnick explained. "This helps explainwhy we see unintended effects of drugs, and opens up a new paradigm for how onehas to think about discovering drugs."

 

 

Proteins' binding pockets are formed by the secondary natureof amino acids, which is directed by hydrogen bonding and results in theformation of similar pockets on many different proteins, even unrelated ones.

 

If the geometry of these proteins and the location of theirbinding pockets could be modeled, drug developers could have new insight intohow small-molecule drugs need to be shaped in order to avoid the worst possibleside effects, and Skolnick says they are already working on such modeling.

 

 

"You can produce models of acceptable quality where you cando virtual screening for effectively three-quarters of any proteome," he says.

 

 

It is possible to look through proteins and construct anetwork, "like a promiscuity index," says Skolnick, to get a rough idea of thesites at which a molecule might bind. Along that same line, once these pocketsare better understood and identified, it could also be possible to shape asmall molecule in order to choose which side effects will result, shaping it toavoid binding with off-target sites associated with important biologicalprocesses.

 

 

"The key next question is to try to move this to the nextlevel, to try to correlate patterns and targets with physiological response tobe predictive…you bind and would like to be able to predict or suggest with abetter than reasonable probability, which is probably realistic at this point,that this collection of targets are likely to have a positive physiologicalresponse," says Skolnick, adding that "one has to think of it as a networkresponse rather than a single molecule response."

 

 

"How do you pick a target or the targets that are going togive you the response that you want? That's where we're heading," he adds.