Combi-Chem

4f. Compound selection procedures

Identification of outliers

The functionality to identify and remove outliers in property space is accessed by going to the LIBRARY ANALYSIS card in the COMBI-CHEM I deck and selecting the Property Space Outliers menu item. This opens the Property Space Outliers control panel.

Outliers are color-coded in the corresponding 3D plot, which is automatically updated. Options are included to display only outlier rows in the study table and to either remove the outliers from the Observations group or to permanently delete the corresponding rows from the study table.

Selecting diverse compounds

Cluster-based selections

To obtain cluster-based selections, the number of compounds desired should be entered as the number of clusters in the Statistical Method Preferences panel.

On the Study Table, select Preferences/Statistical Method to bring up the Statistical Method Preferences control panel. Set Statistical Method to CLUSTER and Method to HCA/Average Linkage.

Clicking the Get Clusters button will intersect the dendogram so as to provide the desired number of clusters. In the case of classical descriptor or principal components, the compound closest to each cluster centroid will be identified and selected as a representant.

When working with fingerprints (ISIS or Daylight), cluster centroids are ill-defined and no compound selection will be performed. However, the Get Clusters procedure will assign a cluster ID to each compound as a new column in the Study Table. There are two work-arounds to this situation:

1.   MDS can be applied to the fingerprints prior to the cluster analysis. The MDS coordinates can then be used as independent variables in the cluster analysis in place of the fingerprints.

2.   The new Cluster ID column created in the Study Table can be used as an independent variable for a cell-based selection. The number of cells specified in the cell-based selection should equal the number of clusters obtained in the previous step. The procedure is equivalent to selecting one compound randomly from each cluster.

Zero-iteration relocation clustering is a very fast clustering method, which assigns each molecule to the cluster belonging to the closest seed molecule. The set of seeds is selected as in standard Relocation clustering: randomly or by selecting study table rows. You may also specify the number of relocations to be performed. Standard relocation clustering corresponds to an "unlimited" number of relocations (i.e., the process continues until the relocations stabilize), and zero-relocation clustering is the opposite: no relocations are performed.

The zero-relocation method is very sensitive to the selection of seeds.

Distance-based selections

Distance-based diversity selection is accessed by going to the LIBRARY ANALYSIS card in the COMBI-CHEM I deck and selecting the Select Molecules Diverse Distance-based menu item. This opens the Select Diverse control panel

You can control:

• The number of compounds to pick.

• The diversity function to use (MaxMin, Product, PowerSum, D-Optimal design, or Max MinSpanTree).

• Which overall function is optimized (Only Diversity or Diversity-Penalty)

• Whether to add to a previously selected set of diverse models, by checking (or unchecking) the Add selected molecules to subset check box. You can use this functionality to incrementally add to the subset of selected models.

• Whether the selection should operate on the complete set of models in the study table or only on models from a specific library (previously defined using the Define Libraries control panel).

• To fine-tune other settings for the diversity selection experiment (click the Preferences... push button to open the Analysis Preferences control panel).

Note

Cell-based selections

The number of cells generated depends on the number of models to select and on the number of properties defined as independent variables. The number of cells is always less than or equal to the number of models to select. Properties are binned with a bias towards the ones that exhibit the largest variation, i.e., those properties with the largest variation (range) tend to have more divisions than the ones that do not vary much for the set of models under consideration.

Cell-based diversity selection is accessed from the Cell-Based Selection control panel, which is opened by selecting the Select Molecules Diverse Cell-based menu item from the LIBRARY ANALYSIS card in the COMBI-CHEM I deck.

You can control:

• Define the binning (Define Binning..., see below). We recommend you adjust the binning of descriptor space so that the number of occupied cells comes close (in default or excess depending on your requirements) to the number of desired compounds.

• The number of molecules to select from each filled cell (Number of Molecules per Occupied Cell).

• The normalization of the columns to be used in the 3D plot (Normalization).

• Which overall function is optimized (Only Diversity or Diversity-Penalty).

• Whether to generate a 3D plot for visualizing the cells and models (Plot Cells in 3D Space).

• Set Plot Preferences (see below).

• Whether to create a Cerius² table and/or an ASCII file with results from the binning of property space, including the number of models in each cell and the model selected (Output Cells Info to Table, Output Results to File).

• Whether to create a column with the cell number for each molecule in the Study Table (Create Cell Number Column).

• Whether the selection operates on the complete set of models in the study table (All Molecules in Study Table) or only on models from a specific library (previously specified with the Define Library control panel) (Molecules in Library).

The Plot Preferences push button opens the Cells Plot Preferences control panel, which allows you to adjust plotting options.

You can control:

• Which molecules to plot (All, Selected, Unselected, or None).

• Which symbols to use for the models (Number, Name, Star, or Dot).

• Which cells to plot (All, Filled, Empty, or one Specific cell).

• Whether to Plot cell numbers.

• Which three independent variables to use as the x, y, and z axes (the three with the largest variation or the three currently selected in the table).

• How to color the models (by selection or by library number).

Note

Selecting similar models

The Select Similar control panel enables you to perform similarity selections around a lead compound. Open the Select Similar control panel by selecting the Select Molecules Similar menu item from the LIBRARY ANALYSIS card in the COMBI-CHEM I deck.

When N Closest Molecules is selected under Similarity Criteria, the Select Similar control panel allows you to:

• Set the Number of models to be retrieved, when N Closest Molecules is selected under Similarity Criteria.

• Choose which overall function is optimized (Only Diversity or Diversity-Penalty).

• Specify to look for all models within a Maximum distance from the original lead.

Selecting random compounds

Random selections may be obtained by using the Select Diverse panel (Distance-based selections) and providing the number of desired random samples as the number of molecules to be selected.

To obtain a random selection, go to the Analysis Preferences panel (click the Preferences button on the LIBRARY ANALYSIS card) and make sure that:

• The Bin Property Space for Initial Selection option is deactivated.

• The number of Monte Carlo Steps is set to 0.

• The Distance Metric popup provides three options for calculating the distance between models in property space:

  City Block:

  Euclidean:

  Canberra:

• The Similarity Coefficient popup sets an algorithm to calculate the fingerprint similarity coefficient (Tanimoto, Dice, Hamming, Hamming sqrt, Ochiai (cosine), Ochiai sqrt)

• The Normalization popup provides five options for normalizing the data (Mean, Variance, Mean/Variance Unit Range, and None,).

• The Plot Function vs. Monte Carlo Step check box specifies whether to generate a plot of the diversity function vs. Monte Carlo step.

• The Update 3D Plot check box specifies whether to update (or generate) the 3D plot for visualizing the diversity or similarity selection.

• The Monte Carlo Steps text box specifies the number of steps for the Monte Carlo Optimization.

• The Monte Carlo Temperature text box specifies the temperature (in Kelvins) for the Monte Carlo Optimization.

• The Monte Carlo Seed text box specifies the seed for the random number generator in the Monte Carlo Optimization.

• The Terminate After Idle Steps text box specifies the number of unsuccessful steps required to terminate optimization.

• The Set Default Values action button returns the Monte Carlo text boxes to their default values.

• The Bin Property Space for Initial Selection check box specifies whether to select the initial set of models in a distance-based diversity optimization run by binning property space and selecting models from the filled cells (this can accelerate convergence of the stochastic optimization run).

• The Gaussian Alpha entry box controls the behavior of the error function used to calculate the Max Min SpanTree diversity function.

• The Bucket Size for K-d Tree entry box specifies the maximum number of molecules in a bucket (terminal node) for building the K-d tree.

• The Number of Nearest Neighbors entry box specifies the number of nearest neighbors to store for calculating the Minimum Spanning Tree.

• The Show Tree in 3D Plot check box specifies whether to plot the minimum spanning tree for the final set of selected diverse models.

• The Show Row Numbers in Tree Plot check box specifies whether to include molecule numbers in the Minimum Spanning Tree.

• The Display similarity matrix check box controls whether to create a table with the similarity coefficients for all pairs of models when doing a Select Similar calculation (this can be very time-consuming for large sets of models).