Maybe the following questions were already asked before but at least i was not able to find a corresponding thread using the search function.

I am new to rosetta@home which is in fact an interesting project especially for me because i am a software engineer for medical applications and so have a rough idea of what is going on in the software. Anyway i am not really familiar with the topic "protein folding", neither in chemical nor in technical way. Because of this, there are a few different questions i have concerning this topic:

1.: As i can see, the project tries to predict the (initially known) structure of a protein chain by randomly folding the amino acid elements given in sequence. In the "best predictions"-section there are some graphics showing structures having a much lower energy than the ones with the lowest RMSD offset. As far as i know, organic structures do not always tend to take up the lowest energetic state, unsaturated fatty acids for example.

My question at this point is: Is the folding of a (really detected) protein structure always the one with the lowest possible energy value? Why do the structures closest to the "original" often offer a higher energy value than the lowest-energy-structure with a much higher RMSD value?

2.: As i read in another thread, the software is able to calculate the overall energy value for a complete protein structure from calculating the binding powers between the several amino acids taking into account the relative angle between the chain elements. If there is a way to calculate the binding power for any given angle wouldn't it also be possible to reverse the whole calculation to determine the necessary angle using a limes function? In this case it would not be necessary to randomly create a model with a HIGH number of random elements that can almost never be exactly the given structure because of the mathematical probability to find the exact structure (even with a specified fault tolerance).

As i said, i am not familiar with the topic and i assume, you guys know what you are doing, i just try to understand why things work as they do and hope you can help me understand the world (at least a little piece more) :)

A.B.

Good questions. I will defer to the Project Team for specific answers, but I believe the lowest energy is the best predictor. It isn't always a perfect match to the native structure, but it is the best predictor available.

From my experience playing with the beta version of the game (read more) I can tell you that the best answer will not have the best energy level between each and every pair of amino acids. It is a compromise which allows the whole thing to stay together. It isn't always able to fold in such a perfect shape. The perfect energy level might fold it right back into another part of itself for example.

It is sort of like a mouse in a maze. If you score the mouse by his distance from the exit, he may find himself with a very high score, yet not be able to see the exit, and in fact may be staring at a deadend. He may have to backtrack a considerable distance in order to ever find the exit. The fact that you see the mouse backtracking should not be taken as randomly wondering through the maze. He's got a lot of experience with mazes and has a method to how he solves them. As the protein chain is constructed, the sidechains and backbone get in the way of maximum energy levels at each link in the backbone.

Dr. Baker has referred to the math problem as being how to find a global minima (of a formula with many codependant factors). (my wording in parenthesis). The neat thing about the game is that all the math is built in to how it reacts to your poking and proding, so you don't have to really deal with that.

---

Rosetta Moderator: Mod.Sense

Thank you for the answer but i'm afraid there is something wrong with it:

The fact that you see the mouse backtracking should not be taken as randomly wondering through the maze.

As a matter of fact, the system does rotate and fold the chains randomly as explained by David Baker in the thread "Cheating ??".

For this reason i was really wondering if a randomly generated structure (although after the rough rendering there is a refinement applied which only varies the angles again randomly but slightly ) can ever be an adequate representation of a natural original. Adequate here means "suitable for scientific needs".

Of course the randomizing is not applied without sense but anyway it is just first generating, then testing the energy. In your example th mouse in the maze would not turn anywhere randomly, walk a little bit and then test its distance to the exit. If it had experience in mazes it would be able to know where to go BEFORE going there (that is what i wanted to suggest to use the "reverse method").

I never know how much reading to presume a new participant has done. I think we are all on the same page here. Here is a link to the post Dr. Baker made. He was specifically discussing the barcode type of search that Rosetta is capable of.

I just want to be sure you understand the difference between the random number seed which is the basis for beginning each model, and randomly wondering through the entire search space. It sounds like you understand that.

Yes, the methods used are already able to provide predicted models with sufficient accuracy to be useful to creating drugs. Indeed, some studies of HIV viruses were done, and a list of potential drug candidates submitted. The continuing research is on how to extend the concepts to get accurate predictions on more complex proteins, along with the studies of RNA, docking, amyloid fibrils and etc.

To further try to repair my potentially flawed mouse and maze analogy... an experienced mouse does not know for certain which way to turn in the maze. Only if his experience is with with the same maze repeatedly could he know for certain. But, his experience gives him a method that he has found to work reliably over a series of different mazes. So his approach may be to flip a coin at a given intersection, but then remember the choice made the first time, and not retrace that path again if he ends up backtracking to that point. So, randomness where a better decision is not possible. Only to give a specific direction to test. Then, in the case of proteins, try the other direction as well to see if a better answer reveals itself there (unless experience tells you that the initial choice does indeed appear to be the best we are likely to find).

You are going to love the game when it is finished. It basically lets you try out your ideas and see if they work well or not.

---

Rosetta Moderator: Mod.Sense