The SAM-T06 hand predictions use methods similar to SAM_T04 in CASP6 and the SAM-T02 method in CASP5. We start with a fully automated method (implemented as the SAM_T06 server): Use the SAM-T2K and SAM-T04 methods for finding homologs of the target and aligning them. The hand method also uses the experimental new SAM-T06 alignment method, which we hope is both more sensitive and lass prone to contamination by unrelated sequences. Make local structure predictions using neural nets and the multiple alignments. We currently use 10 local-structure alphabets: DSSP STRIDE STR2 an extended version of DSSP that splits the beta strands into multiple classes (parallel/antiparallel/mixed, edge/center) ALPHA an discretization of the alpha torsion angle: CA(i-i), CA(i), CA(i+1), CA(i+2) BYS a discretization of Ramachandran plots, due to Bystroff CB_burial_14_7 a 7-state discretization of the number of C_beta atoms in a 14 Angstrom radius sphere around the C_beta. near-backbone-11 an 11-state discretization of the number of residues (represented by near-backbone points) in a 9.65 Angstrom radius sphere around the sidechain proxy spot for the residue. DSSP_EHL2 CASP's collapse of the DSSP alphabet DSSP_EHL2 is not predicted directly by a neural net, but is computed as a weighted average of the other backbone alphabet predictions. O_NOTOR2 an alphabet for predicting characteristics of hydrogen bonds from the carbonyl oxygen N_NOTOR2 an alphabet for predicting characteristics of hydrogen bonds from the amide nitrogen We hope to add more networks for other alphabets over the summer. We make 2-track HMMs with each alphabet (1.0 amino acid + 0.3 local structure) and use them to score a template library of about 8000 (t06), 10000 (t04), or 15000 (t2k) templates. The template libraries are expanded weekly, but old template HMMs are not rebuilt. We also used a single-track HMM to score not just the template library, but a non-redundant copy of the entire PDB. One-track HMMs built from the template library multiple alignments were used to score the target sequence. All the logs of e-values were combined in a weighted average (with rather arbitrary weights, since we still have not taken the time to optimize them), and the best templates ranked. Alignments of the target to the top templates were made using several different alignment methods (mainly using the SAM hmmscore program, but a few alignments were made with Bob Edgar's MUSCLE profile-profile aligner). Generate fragments (short 9-residue alignments for each position) using SAM's "fragfinder" program and the 3-track HMM which tested best for alignment. Residue-residue contact predictions are made using mutual information, pairwise contact potentials, joint entropy, and other signals combined by a neural net. The contact prediction method is expected to evolve over the summer, as new features are selected and new networks trained. Then the "undertaker" program (named because it optimizes burial) is used to try to combine the alignments and the fragments into a consistent 3D model. No single alignment or parent template was used as a frozen core, though in many cases one had much more influence than the others. The alignment scores were not passed to undertaker, but were used only to pick the set of alignments and fragments that undertaker would see. Helix and strand constraints generated from the secondary-structure predictions are passed to undertaker to use in the cost function, as are the residue-residue contact prediction. One important change in this server over previous methods is that sheet constraints are extracted from the top few alignments and passed to undertaker. After the automatic prediction is done, we examine it by hand and try to fix any flaws that we see. This generally involves rerunning undertaker with new cost functions, increasing the weights for features we want to see and decreasing the weights where we think the optimization has gone overboard. Sometimes we will add new templates or remove ones that we think are misleading the optimization process. New this year, we are also occasionally using ProteinShop to manipulate proteins by hand, to produce starting points for undertaker optimization. We expect this to be most useful in new-fold all-alpha proteins, where undertaker often gets trapped in poor local minima by extending helices too far. Another new trick is to optimize models with gromacs to knock them out of a local minimum. The gromacs optimization does terrible things to the model (messing up sidechains and peptide planes), but is good at removing clashes. The resulting models are only a small distance from the pre-optimization models, but score much worse with the undertaker cost functions, so undertaker can move them more freely than models it has optimized itself. For this model, we had two strong hits to the PDB and to our HMM's, 2g03A and 2f6sA. The best hit PDB hit to 2g03A matched residues 65-223 of the target. However, since the target is 299 residues, we needed to split the model into domains for the first region (residues 1-65) and the last region (residues 223-299). These two regions had no close PDB hits and poor fold recognition hits, so we had to model these using ab initio techniques. For almost all the models we made chimeras of the three domains in order to make decent models, and we had to include constraints to keep the seven helix bundle for the comparative modelling portion of the protein together. Model 1 is try10-opt2. This is a chimera of try4, which was the first polish of the best SAM_T06 server model, and using the first run of our subdomain from L225-T299. We included residues 1-229 of the polished SAM_T06 model and residues 230-299 of the subdomain we ran. This model scores the best using the rosetta scoring function. It also scores best using the try10 costfcn, which is a fairly neutral function that we decided to use. Model 2 is try12-opt2. This model is also built from a chimera. The chimera was built by using three separate predictions from subdomains. The first subdomain was from M1-D64, and used a model based on secondary structure prediction and hydrogen bond predictions in the first subdomain. Residues T65-R224 used try3-opt2, which was a polished model built from alignments and using distance constraints to keep the seven helix bundle together. The alignments and constraints were from 2g03A. The third domain was from L225-T299 and used try1-opt2 from this region, which was the original results from our first run in this subdomain. It scored well on two costfcns that we used to analyze the models. It scored first on the unconstrained costfcn and scored second best on the try10 costfcn. Model 3 is try13-opt2. This model is a polish of the complete best scoring server model from the SAM_T06 server. It kept the seven helix bundle in the center of the protein and model the rest of the protein fairly nicely. Model 4 is try20-opt2. This model is a polish of another chimera that we had built. This chimera was also built from three subdomains. M1-T64 used try4-opt2, which was the model based on secondary structure predictions. A65-L224 used try4-opt2 which was another model built with constraints on helices and alignments, but this time using constraints from the other well scoring model, 2f6sA. L225-T299 used try6-opt2, which attempted to use predicted residue-residue contacts, predicted hydrogen bonds, and predicted secondary structure. This chimera was optimized and polished in undertaker. This model scored second best on the unconstrained costfcn and scored decently on the try10 costfcn. Model 5 is try17-opt2.gromacs0.repack-nonPC. This model scores very well with rosetta. I included it here for a bit of variety for submissions. This model was also built from a chimera. This model used try2-opt2 of the M1-T64 subdomain, which was a polish of the initial undertaker run for this domain. A65-L224 used try4-opt2 which was based on 2f6sA. L225-T299 used try5-opt2 of this subdomain which attempted to used secondary structure predictions and hydrogen bond predictions for this domain. This model was a polished model of the chimera.