From chemistry-request@ccl.net Thu Jan 24 13:36:41 1991 Date: Thu, 24 Jan 91 11:11:12 CST From: berkley@wubs.wustl.edu (Berkley Shands) To: chemistry@ccl.net Subject: Sybyl OMAP tips Status: RO Determining a receptor geometry by searching a number of related compounds used to be a slow and very tedious act. Prone to systematic errors and constrained by time, a limited number of experiments are used to deduce the receptor's shape. This is no longer the case; There are a few practical steps that can be taken to avoid excessive computation. Regardless of the search program being used (I'll assume SYBYL from TRIPOS Associates for now), there are some easy ways to reduce your exposure. 1) run molecules in increasing order of rotatable bonds. start with the most constrained within the class; 3 rotatable bonds first, then 4, then ... This will give you the least runtime to produce initial results. Thereafter, you can run searches in parallel. 2) Ignore side chains, and other features not central to the geometry you are looking at on the first pass. Whatever results you obtain must be a superset of the end product. you can then use the results as a constraint on later attempts. 3) Do not output angle files, energies, or any other such data until you have determined the results are final. A million angle file entries will not be looked at :-). 4) Use a fairly large grid size (0.4, 0.25) for matching OMAP points. Assuming that the geometry will lie within one grid spacing ignores Murphy's law. Our research has shown that some dimensions of the geometry will never line up on a regular grid. Either you move the grid (hard) or you make the grid larger. 5) If you can afford to, make a trial run of all molecules at 20 to 30 degrees, unconstrained by a previous run. Sort the molecules based on rotatable bond count, then OMAP points produced. This is usually within 20% of the fastest order obtainable. 6) Monitor your searches, if you have found all OMAP points, abort the search. Why continue looking for more conformations? This can gain you much time. 7) Use a prime scan factor when searching. The union of several searches at differing scan factors closely approximates a single scan at a much lower scan factor. It will save much time as well. If your runs do not agree in their results, you must use a lower scan factor because you have just discovered you missed an important chunk of the search space. 8) Examine the OMAP points to see if there are any groupings. Similar points may be considered as one large point. Scattered points may indicate the scan factor is too large. For an example of #4, consider a fictitious series called RAT. RAT is 5 molecules long, and is not very flexible. A two dimensional grid is used, with a 0.20 Angstrom grid size. Molecule #1 produces an OMAP point at (1.357,2.156). Molecule #2 generates a point (1.405, 2.206). Molecule #3 generates (1.394,2.199). Molecule #4 generates (1.411, 2.210) and molecule #5 generates (1.450, 2.144). When rounded to 0.2 angstroms, you will get different OMAP points. However, the size of the geometry indicates 0.2 is reasonable. Since the "points" do not line up, you are forced to use a larger grid size to make the points fall into the same grid slot. Using a smaller scan factor may help, but it will be costly in terms of time spent searching. As an example of #7, consider our results with the ACE series, using only 28 molecules. Runtime Summary (IRIS 4d/380) Scan Points Groups Processor Elapsed Comments Factor Found Found Time Time ====== ====== ====== ========= ======= ============= 15 0 0 1 47 Last molecule steric bumped 14 1 1 1 51 13 1 1 4 41 12 1 1 4 46 11 4 1 5 47 10 3 1 7 41 9 5 1 16 51 8 8 1 37 67 7 31 6 106 79 6 22 3 150 98 5 12 2 206 163 4 45 5 1482 506 3 39 4 4094 1922 The union of the 4, 5 and 6 degree scans produced over 60 points, in 4 groups. The two degree scan produced 70+ points, in four groups. A "group" is a collection of points that share physical proximity. These runs made use of a multiprocessed search program, using 4 processors, reported in seconds. Berkley Shands berkley@wucs1.wustl.edu ---