Posts by phil

1) Message boards : Rosetta@home Science : amyloid fibril structure prediction (Message 31220) Posted 16 Nov 2006 by phil Post: So you start with Abeta(10-40) which means 30 structures. When do you think you will have established those 30 structures and what's the plan afterwards? Do you have a definite set of amyloids you want to study or does it all depend on the progress you make? How accurate do you want to predict the structure? I understand high-resolutions means pretty accurate but is you goal a prediction as accurate as X-ray crystallography (<1.5 Angstrom)? I didn't do a good job of explaining my terminology: when I say Abeta(x-y), x and y are positions in the protein sequence. eg 10-40 means residues 10-40 of the abeta peptide, or 31 amino acids. The plan is hopefully to propose further experiments and iterate between experiment and simulation to converge on a "high-resolution" structure, at least for the core regions. I think 1.5 Angstroms would be fantastic -- in reality some parts of the structure may be disordered and these we just have to ignore.
2) Message boards : Rosetta@home Science : amyloid fibril structure prediction (Message 31219) Posted 16 Nov 2006 by phil Post: ...If we have a three-dimensional model then we can predict what this [x-ray diffraction] pattern should look like and compare with the pattern that is actually observed. ...and presumably we (mankind) can learn to do the opposite? Use what is known from the diffraction pattern to bias the computational modelling? (thus allowing the computational model to benefit from the x-ray studies, and arrive at the correct answer with less computing power expended) ...or is this already done to some extent. This is done in the case of x-ray diffraction patterns taken from crystals. It's more challenging to derive detailed structures from fiber diffraction data but perhaps this could be done. But you can see features in the data that correspond to structural repeats -- eg the 4.8A repeat that corresponds to stacked beta-strands.
3) Message boards : Rosetta@home Science : amyloid fibril structure prediction (Message 31218) Posted 16 Nov 2006 by phil Post: What interests me is the question of water in the structure I know the general Rosetta sim is based on a single protein surrounded by water, and if I understand correctly the model caters for the different enrgy regime where water is present (outside the structure) and where it is absent (inside folds which are tight enough to exclude water), so that the modelled protein follows the real one in folding up the hydrophobics on the inside away from the water. My speculation is that there may be pockets of water trapped between the identical copies of the protein/peptide, for example sandwiched between two hydrophilic amino acids. In contrast to the normal Rosetta case, where the positions of the water molecules are not constrained outside the protein, but only by the protein itself, the water molecule may even be fixed at both ends. First does this make sense in biochem terms? Second, if so, how does your model approach all the extra degrees of freedom that follow from having the position of the relatively tiny water molecules becoming significant? River~~ This is a very important question. In fact I've read a number of papers which argue the same thing, that there may be columns of water inside the fibril. For most applications we use an implicit solvation model, which tries to assess the degree to which each atom is accessible to water by summing contributions from all nearby atoms. Polar atoms prefer to remain solvent accessible while hydrophobic ones will prefer to be buried. This does not distinguish between the inside and the outside of the fibril per se -- if there is a void in the center of a model the solvation term will reward polar atoms that neighbor the void on the assumption that water could be there. Unfortunately this doesn't take into account the fact that water has a discrete size and some voids may not be large enough. We do also have explicit water models but I am not using these for the fibril modeling because they are too computationally expensive. In summary you may be right that internal molecules are present and it may be necessary to model them explicitly, which would indeed increase the number of degrees of freedom... I am excited to analyze the models that we are getting to look for internal cavities.
4) Message boards : Rosetta@home Science : amyloid fibril structure prediction (Message 31131) Posted 14 Nov 2006 by phil Post: Could to help us to understand what "medically relevant" really means to you? And once you feel you've got a good model of this... how will you test to determine if you're correct? Or close? given that there is no crystalization experiment to compare to? Will you compare with NMR data? Or test for interactions with other known structures? Good questions. How will we validate our model? Mainly by comparison with experimental data. We can propose new solid-state NMR experiments that would support or contradict the model, eg, if we say that the sidechains of residues 19 and 34 are within X Angstroms then they can do an experiment with peptide labeled at those positions and see if they get a signal. In addition, there are a variety of experiments whose results one could predict from a structural model. For example, even though the fibrils don't form ordered 3-dimensional crystals suitable for X-ray crystallography, they do have enough 1-dimensional order to yield characteristic diffraction patterns when you shoot X-rays through them. If we have a three-dimensional model then we can predict what this pattern should look like and compare with the pattern that is actually observed. THere are several other experiments of this type which can't be used to determine the structure but which can support or refute an existing model. As for "medically relevant" -- I confess I have very little clinical background or experience. My naive definition is to be associated with one or more human diseases. Thanks for the interest and please do post with more questions which I may or may not be able to answer! Take care, Phil
5) Message boards : Rosetta@home Science : amyloid fibril structure prediction (Message 31106) Posted 14 Nov 2006 by phil Post: Hi guys. Thanks for the posts. I too am excited about harnessing the amazing computational power that you are donating to improve our understanding of the basis of amyloid fibril formation. Here's a bit more background. Rather than one giant post I'm going to reply to the thread multiple times as I find useful links and other information. Amyloid disease: a number of diseases (Alzheimer's, Parkinson's, Type II diabetes, the transmissible spongiform encephalies "Mad Cow",...) are associated with the deposition of protein aggregates in the diseased tissues. It's been very difficult to determine the molecular structure of these aggregates because they are fibrous and insoluble, resisting crystallization. For this reason they are attractive targets for computational modeling. We are working on building high-resolution predictions for medically relevant amyloids, starting with the Alzheimer's-beta fragment (1-40). We are collaborating with a group at the NIH who use solid-state NMR to probe the structure of Abeta1-40 aggregates. Our goal is to determine by simulation the set of low-energy conformations that are compatible with the current collection of experimental distance constraints that they have given us. By analyzing this set of solutions we will propose new experiments which should be maximally informative in determining the structure. We will then use the results of these new experiments to build models and iterate this procedure to generate high-resolution models. Why predict amyloid structures? The expectation is that knowledge of the high-resolution structure of amyloid fibrils will aid in the design of small molecules which can inhibit fibril formation and/or break down existing deposits. The simulation: fibrils are thought to be symmetric assemblies, which means that they are formed by stacking together many identical copies of a protein or peptide. What you see on your screen are 12 copies of the Abeta(10-40) sequence (our collaborators tell us that residues 1-9 are likely to be disordered and outside the core of the fibril). There will be a low-resolution simulation (no sidechains visible, lots of movement), followed by a high-resolution refinement (sidechains visible in the left panel, simulation appears to go much more slowly). Not all low-resolution trajectories will go on to high-res refinement -- we are filtering for conformations which satisfy the experimental constraints. Wikipedia entry for Amyloid OK -- post with questions and I will continue to look for info to add. Thanks for your interest and your spare cycles!! -Phil
6) Message boards : Rosetta@home Science : amyloid fibril structure prediction (Message 31077) Posted 13 Nov 2006 by phil Post: This thread will feature discussion of a new type of simulation being introduced in the 5.40 release. We are trying to build high-resolution models of the fibrils formed by medically-relevant amyloids. The first prediction jobs will focus on the Alzheimer's beta(1-40) fragment. More details to follow.