Pipeline Pilot

The pragmatic beauty of the chemical fingerprint is that the more common features of two molecules that there are, the more common bits are set. The mathematic approach used to translate the fingerprint comparison data into a measure of similarity tunes the molecular comparison [5]. The Tanimoto similarity index works well when a relatively sparse fingerprint is used and when the molecules to be compared are broadly comparable in size and complexity [5]. If the nature of the molecules or the comparison desired is not adequately met by the Tanimoto index, multiple other indices are available to the researcher. For example, the Daylight software offers the user over ten similarity metrics, and the Pipeline Pilot as distributed offers at least three. Some of these metrics (e.g., Tversky, Cosine) offer better behavior if the query molecule is significantly smaller than the molecule compared to it. 

Figure 6. Distributions of essential computed molecular properties defining drug-likeness for selected compound sets. Shown are the fraction of compounds vs. the properties. Orange NIBR historical medicinal chemistry collection. Brown Compilation of combinatorial chemistry libraries. Dark Green Drugs (launched or Phase III listed in MDDR or CMC). Brown Compilation from combinatorial libraries. Pink Natural products of DNP. tight Green HTS hits of NIBR 2004 screens. All properties were calculated with Pipeline Pilot software www.scitegic.com).

Figure 7. Multiclass NaTve Bayes modeling within Pipeline Pilot software (www.scitegic.com) based on the WOMBAT chemogenomics dataset Probabilistic target predictions are possible for compounds given only their chemical structure. In the example shown, the WOMBAT targets were predicted for Calphostin C, a known protein kinase C inhibiting natural product Tubulin and beta-hexosaminidase are predicted as additional possible targets.

Before Pipeline Pilot, retrieval of assay data (for us) was done mainly via RS3 for Excel (5). RS3 allows one to obtain information on a given assay or a set of assays with queries that are set up by clicking on the folders containing... 

As an alternative, we have made extensive use of the data retrieval and manipulation functions within Pipeline Pilot to provide more powerful querying capabilities without sacrificing ease of use. Figure 4 shows the user interface for the standard ad hoc assay protocol used by most chemists and... 

While the protocol described above has proven very useful in its own right, it often serves as the first step in designing custom protocols for a team. Once a user is satisfied with the results returned by the ad hoc query, the next step is often to hardwire these queries into a protocol that provides one-button access to all the pertinent data for a project. Figure 5 shows such a protocol used to retrieve data for several assays in a Neuroscience project. This protocol highlights an additional filtering option that we make use of in many of our protocols. The text box ( selected cmpds from list ) (7) supports the input of a list of identification numbers (in our case either jnjnumber or batchid ). This allows data to be retrieved on select compounds as opposed to all those tested in the assays. This feature is very popular and involves the use of Perl within the Pipeline Pilot protocol. All this work is done on the server and the results are written to a SD file that can be downloaded to the desktop. At this stage either DIVA or Accord for Excel is typically used to view the files. 

We have illustrated the capability of Pipeline Pilot to query all the data on a set of compounds from the internal database and perform simple filtering on them. Additionally, the ability of Pipeline Pilot to easily access other (several) databases (or files) besides the internal database provides us with a very powerful tool for data mining. The utility of accessing multiple databases simultaneously is discussed in the following sections. Here we give an example of how being able to access different databases, other than the in-house one, and perform sophisticated filtering on an the data is in itself a tremendous asset. 

Fig. 6. Pipeline Pilot protocol used to filter reactants useful in alkylation.

Fig. 7. User interface of the Pipeline Pilot similarity search protocol.

Figure 7 shows the web interface for our Pipeline Pilot-based similarity search engine. For similarity searching a connectivity fingerprint (available within the Pipeline Pilot software) is used and the Tanimoto coefficient is calculated. In the text box (labeled JNJNumberList under Parameters ) one can type or paste (e.g., from an Excel sheet) a list of identification numbers ( jnjnumber, batchid ) to be used as probes for the search. Alternatively, an... 

SD structure file or a file containing a list of identification numbers [Excel or Comma Separated Value (CSV) format] can be uploaded. In the Excel or CSV cases the protocol will automatically look up the structure that corresponds to the identification number, as it does for the text box input. A useful feature (see Databases under Parameters in Fig. 7) is the ability to search multiple databases in a single search. This is made possible by the use of Perl in the Pipeline Pilot protocol. The Perl code is very general and easily allows for the addition of new databases as they become available, thereby further increasing the versatility of this protocol. 

Both DIVA and RS3 provide some functionality in terms of substructure searches (SSS), although it is somewhat limited. For example, DIVA searches can only be performed on data that have already been queried from the database ). This pre-queried data need to be readily available to DIVA either via RS3 or as an SD file. In the case of RS3, the inclusion of multiple data sources (e.g., searching the corporate database and an external vendor library) is not trivial. As a result, while DIVA and RS3 are very useful for SSS under certain conditions, they are not as robust when compared to the Pipeline Pilot protocol. 

Like the similarity protocol, the user interface for the Pipeline Pilot SSS protocol supports multiple databases and multiple probes in a single query. The input of the probe molecule(s) is accomplished using a SD file, which can be generated in most standard chemical drawing programs. This file is supplied as one of the inputs in the SSS interface. The use of identification numbers is not as applicable here as it is in the case of the similarity protocol since we are most likely not using existing molecules as substructures. Consequently, textbox input is not an option nor is the upload of a list of numbers in CSV or Excel format. 

Fig. 11. Using the generic web services interface of Pipeline Pilot it is possible to generate complex input forms. In this case we use different variable types to provide a Vendor Filter protocol for selecting reactants.

At this stage several different algorithms could be applied to filter the reactant list (17,18). The simple example we use here allows one to filter reactants based on preferences for vendors. This filter is easy to implement using the Pipeline Pilot web interface (Fig. 11). Using this filter, compounds only available from vendors in the vendor reject list are removed. The remaining compounds are then sorted so that reactants from preferred vendors are listed first. 

The textbox feature discussed in the similarity and substructure protocols and appearing in many other protocols not mention herein is due to an interfacing of Perl code with Pipeline Pilot written by Mike Hack in the J J PRD La Jolla CADD group. 

The pipeline pilot protocol as a well as the initial version of the web interface for the protocol was written by Chris Farmer at Scitegic. The web interface was refined by Andre Volkov from J J PRD. 

In order to classify promiscuous and selective compounds, we used the NB modeling protocol available in Pipeline Pilot (Scitegic) [53]. The data was split randomly into 5193 compounds for modeling and 574 compounds for testing the models. In addition to the test set, 302 known drugs were also profiled and kept separate for testing the models. All sets were checked visually to ensure that no chemical classes were overrepresented in one set or the other. 

Scitegic Pipeline Pilot 6.1, Scitegic Inc., 9665 Chesapeake Drive, Suite 401,... 

Pipeline Pilot distributed by Accelrys Inc. can be used to enumerate libraries defined either by reactions or by Markush structures http //accelrys.com/resource-center/case-studies/enumeration.html, last accessed February, 2010. 

Since LEAP1 was built based on Pipeline Pilot technology, multiple molecular fingerprints and similarity methods can be applied at disposal, which currently include MDL Public Keys and different levels of FCFPs and ECFPs (18). 

Pipeline Pilot from SciTegic http //www. scitegic.com/... 

Fig. 15.8. Substructure mapping, highlighting, and drill-down. Based on on-the-fly substructure query and mapping capability within SciTegic Pipeline Pilot, PGVL Hub allows user to perform substructure queries into a set of target molecules. In the example shown, a set of substructure queries globally collected and validated as undesirable substructure features to be avoided are mapped into target molecules (41b).

---