| 查看: 979 | 回复: 11 | |||
| 当前主题已经存档。 | |||
| 当前只显示满足指定条件的回帖，点击这里查看本话题的所有回帖 | |||
[交流]
【求助】Sybyl 8.0 关于QSAR的手册 已有2人参与
|
|||
|
本人想做3D-QSAR ,但是现在用的版本是7.3的，在CoMFA这块对数据的要求没有。听说8.0版本CoMFA对数据的要求已经列出，所以想学习学习。 希望大家帮帮忙，帮我找找QSAR那一块的手册。 谢谢！！ |
回复此楼» 猜你喜欢
遇见不省心的家人很难过
已经有9人回复
-
博士延得我，科研能力直往上蹿
已经有3人回复
-
退学或坚持读
已经有24人回复
-
免疫学博士有名额，速联系
已经有14人回复
-
面上基金申报没有其他的参与者成吗
已经有4人回复
-
多组分精馏求助
已经有6人回复
|
谢谢！！！！！我知道不会差太多，但是我就是想详细了解一下这部分. 在SYBYL8.0的手册中，终于以书面的形式告诉用户CoMFA对数据的要求： 2.5.2 Prequalifying your data Experimental Data Your experimental data should possess two properties: It should cover a range of at least 2-3 orders of magnitude It should be evenly distributed over this range First make sure you are working on a log scale. If your experimental data is a concentration (e.g. on an nM scale, such as KD or IC50 values), take the negative log of that value on an M-1 scale (i.e. to convert the values into pKD or pIC50). Using these new log values, make sure the difference between the minimum value and the maximum value is at least 2 or 3, and that the log values are evenly distributed over the range. Even if the difference is greater than 2, if the log values are heavily grouped in one or two regions, then the experimental data is not adequate for use with CoMFA. When you run PLS, you will use the log value (e.g. the pKD or pIC50 value), rather than the originally measured concentration value, as your dependent column. Structural Data There are a number of important factors to know about your molecules prior to running CoMFA. You should have a minimum of 10, preferably 15 or more, molecules in order to create a good CoMFA model. Likewise, you do not want to have too many molecules in your dataset either. In general, you should not try to create a model using more than 50-75 molecules. CoMFA creates a region around all of the molecules in your dataset. As a result, the size and shape of your model structures are important. If the molecules are all very compact, or all fairly flat, the region created will also be compact, or flat, and the model will have difficulty with accurate predictions for molecules that are not compact, or not flat. Your dataset should include a fairly diverse subset of the structures you plan to predict. If the model structures are all of a particular class, or do not contain a wide variety of functional groups, accurate predictions for structures outside of this class, or containing other functional groups, may be more difficult to achieve. If you start with a ligand conformation from a known crystal structure, then it is fairly easy to create other analogs using that conformation as a starting point. But if such data is not available, then you may need to examine multiple conformations of the same ligand before determining what conformation yields the best CoMFA model. In general it is important to remember that the more different a molecule is from the molecules used to generate the model, the greater the chances are the model will not accurately predict its value. |
3楼2010-03-27 14:18:37
tjegg
铁杆木虫 (著名写手)
- 应助: 26 (小学生)
- 金币: 6869.3
- 红花: 16
- 帖子: 2615
- 在线: 448小时
- 虫号: 338983
- 注册: 2007-04-05
- 性别: GG
- 专业: 抗肿瘤药物药理
2楼2010-03-27 12:25:25
4楼2010-03-27 14:19:59
snoopyzhao
至尊木虫 (职业作家)
- 应助: 157 (高中生)
- 贵宾: 0.02
- 金币: 18844.7
- 红花: 29
- 帖子: 3803
- 在线: 1422.4小时
- 虫号: 183750
- 注册: 2006-02-13
- 专业: 污染生态化学
5楼2010-03-28 08:25:26