admin Staff Lv 1
Hi Everyone!
I’m Scott, one of the scientists in the Baker lab. I've been working on computational methods to design hydrogen bond networks, much in the same way as this new puzzle. I’ve looked at some of your solutions, and I’m really, really excited by them! This is the first puzzle, and we are still working out some bugs, but you are already coming up with some nice networks that are “difficult” cases for us to design using our automated methods with Rosetta. I want point out some cool examples and share what we’ve learned so far, as well as describe some details about what makes for “good” networks and designs.
As Jeff mentioned in his great blog post (http://fold.it/portal/node/2000666), it’s important for the networks to be satisfied. A perfectly satisfied network means that every atom that can make a hydrogen bond (indicated by blue and red), actually makes a hydrogen bond. This is important because if an atom can make a hydrogen bond but does not (it’s unsatisfied), it’s likely that in real life this atom will interact with a water molecule, and the protein will deviate from the design (water is a giant hydrogen bond network).
In particular, the following solutions have beautiful networks where nearly everything satisfied:
solution_bid_0000_0000352715_0000987632_0308980440.ir_solution (residues 8+12+59+60+64+67)
solution_bid_0003_0000552390_0000988042_0309024230.ir_solution (residues 60+64+68+9+8+12+59+63+66)
solution_bid_0014_0000263916_0000000000_0308929546.ir_solution (residues 57+60+19+22+52+59+63)
solution_bid_0015_0000450972_0000000000_0308970452.ir_solution (residues 52+53+56+57)
solution_sid_0000_0000104378_0000000000_0309114426.ir_solution
solution_sid_0006_0000447652_0000988042_0308867088.ir_solution
For the methods I am developing, it is hard to exhaustively sample our entire design space and pick out all of the satisfied networks without our computers exploding (figuratively) and taking forever. So we need shortcuts how to get to the “good” solutions faster. That makes it a great problem for FoldIt.
Also, to have a really good designs, it’s important to have a balance of both hydrophobic residues AND hydrogen-bond networks (which are made of polar, hydrophilic residues) — the hydrophobic parts bring things together, and the hydrogen bond networks (the hydrophilic parts) lock them into place in the right position. This is especially true for design puzzles with interfaces, like this symmetric trimer. When this protein is actually made, each of the 3 symmetric parts start out separated and have to “find” each other in a sea of water to assemble; all water must be competed away, and the parts must be driven together in the right orientation.
Hydrophobic literally means “water-fearing” (hydrophobic residues are more attracted to each other than water). And hydrophilic literally means “water-loving” (interacts well with water). This means that the atoms that are making hydrogen bonds in the network can also interact with water; but for the networks to form, they have to out-compete water. So much like cases I described above, where a hydrophilic atom will interact with water if it is not participating in a hydrogen bond, if your design interface is only hydrogen bond networks, it’s unlikely that all of the water will be designed away. Hydrophobic residues act like oil in water, driving these parts of the protein together (we call this the “hydrophobic effect”); however, the hydrophobic residues are not specific — the oil will just stick to any other oil when placed in water. That’s where the hydrogen bond networks come in — they give us specificity, because the hydrogen bonds can only form in specific ways. And that's why we need both: attraction and specificity.
Two solutions that did a nice job of balancing both are solution_bid_0015 and solution_bid_0016. Many solutions I saw had 1 or 2 good networks, but they were surrounded by hdyrophilic residues of worse networks (less satisfied). A better solution would be to keep your best networks and then design hydrophobic residues around them, as opposed to additional networks that are not as good. We are working hard to make the scores and gameplay reflect this.
I look forward to seeing what cool new designs you all come up with!
Scott Boyken
(postdoc, Baker lab)