COMP 254 |
| Quantitative Structure-Activity Relationship (QSAR) modeling has now acquired complex dimensions from its humble beginnings. At times the focus of modelers is on fitting some model equation compromising comprehensibility in the process. For the purpose of interpretable models a two-pronged approach can be followed. One is to use intuitive descriptors in QSAR modeling. Another approach, that is presented here, advocates using simpler models over more complex ones. Model complexity can be defined in terms of algorithm complexity, number of descriptors used in the model, computation time required for training, model interpretability, etc. However, learning better models would obviously be preferred over simpler models. We present an approach wherein the user follows simple flowcharts that guide him in the modeling process, taking him from simple to complex algorithms as and when suitable model cannot be fit on the data. Several experience-based guidelines and technical tips that facilitate QSAR modeling are also presented. |
|
Phil Magee Memorial Symposium: QSAR Reborn
1:30 PM-4:50 PM, Tuesday, August 21, 2007 BCEC -- 156C, Oral
Division of Computers in Chemistry |