E-State RQSAR Panel

In this panel you can create a QSAR model for the activity as a function of the kind of R-group at each attachment position, and display the results qualitatively in terms of increasing the activity, decreasing the activity, or little effect on the activity.

Using the E-State RQSAR Panel

The QSAR model is built using counts of E-state (electrotopological state) atom types present in the R groups at each position as the independent variables. As the effect of hydrogen is absorbed into the attached E-state atom type, the R group includes the attached core atom so that an R group consisting of only a hydrogen can be treated.

A partial-least-squares (PLS) procedure is used to fit the data, assigning 75% of the data to the training set and 25% to the test set at random. The fitting is repeated many times, each time picking the best model that does not overfit the data (which occurs when the standard deviation of the fit is smaller than the error in the property being fit). The mean of the coefficients from all of these trials is used to determine whether a feature at a given position contributes significantly to the property, either positively or negatively: if the absolute value of the mean is greater than the supplied cutoff, the contribution is considered significant.

When the calculation finishes, the attachment positions are labeled with a list of letters representing the E-state atom types, colored by significance: red for significant positive contributions, blue for significant negative contributions, and gray for insignificant contributions. If an E-state atom type is absent from an attachment position, it is not included in the annotation for that position.

To generate a QSAR model:

1. Choose a property from the Property option menu
2. Specify the error in the property in the Error box.
3. Specify the cutoff for significant contributions in the Importance cutoff box.
4. Click Start

E-State RQSAR Panel Features

• Display area
• Property option menu
• Error option menu
• Importance cutoff option menu
• Key to annotations section
• Start button
• Display Details button
• View Atomic Contributions button
• Export CSV button
• Export Image button

Display area

Displays the core, annotated at each attachment point with the E-state atom types and their contribution to the selected property. The annotation consists of letters that represent the E-state atom types, which are defined in the Key to annotations section. Features that were not found at a given attachment point are not represented. Features that have significant negative contributions to the property are colored blue and placed on the left. Features that have significant positive contributions to the property are colored red and placed on the right. Features that have insignificant contributions are colored gray and placed in the middle.

Property option menu

Choose the property used for the activity from this option menu. A message is displayed to the right, indicating how many of the structures analyzed have the selected property.

Error option menu

Enter the estimated error in the values of the property in this box. The value can be derived however you like. It is used to check for overfitting in the PLS procedure: the number of factors used is the largest number at which standard deviation of the regression is larger than the error given in this box.

Importance cutoff option menu

Specify the cutoff on the mean value of the PLS coefficient for an E-state atom to be considered to significantly affect the predicted property. The units of the cutoff are the same as those of the property.

Key to annotations section

The key to the annotations is described in this section. The letters used to represent the atom types are listed with the names of the atom types and the intrinsic E-state value (which increases with increasing effective electronegativity). The atom type names are represented by a string of bond types (s=single, d=double, t=triple, a=aromatic), followed by an element symbol, then a hydrogen count, then a letter that indicates the charge on the atom (p=plus, m=minus). For example, aasC represents a carbon connected to other atoms by two aromatic and one single bond, such as in biphenyl. dssNp represents a positively charged nitrogen bonded with one double bond and two single bonds, such as in an iminium group.

Start button

Start building the QSAR model. When the model is built, the structure is annotated with the E-state atom types, colored according to their contribution to the selected property.

View Atomic Contributions button

View contributions from each atom to the predicted property. Opens the E-State RQSAR - Atomic Contributions panel.

Display Details button

Display a summary of the information obtained from the partial-least-squares procedure. For each E-state atom type, the minimum, maximum, mean, and standard deviation for the PLS coefficient is given, and the assessed effect on the activity is listed, as weak, positive, or negative.

Export CSV button

Export the information displayed in the summary panel to a CSV file. Opens a file selector, in which you can navigate to a location and name the file.

Export Image button

Export an image of the annotated structure in TIFF, JPEG, or PNG format. Opens a file selector, in which you can choose the format and name the file.

• Additional Resources