return to case studies

Chemical Structure Activity Relationships

Client

The client is a combinatorial chemistry vendor of compound libraries. The client is interested in optimizing return on investment in drug design by using leading-edge computational approaches to predict biological activity.

Challenge

The client needed to correlate combinations of structural and physical-chemical properties stored in a large database of small organic molecule and biological activity data. These correlations would enable the client to optimally predict the active molecules with minimal liver toxicity in accordance with the client's pre-determined utility function.

Objectives

AnVil's goal was to provide the client with a prediction of those structural and physical-chemical features that would optimize the client's yield of biologically active and low toxicity molecules.

Solutions

AnVil first applied a variety of supervised and unsupervised machine learning techniques to determine the best combination of structural features to extract from the data in order to optimize the client's utility. After an appropriate structural feature set was selected, AnVil applied its proprietary high-dimensional analysis and visualization tools and processes to aid in understanding both the effect of each structural feature on prediction and how combinations of such features might affect predictive accuracy. The client set the cost of misclassification using a process specific utility function to define an optimal operating point on the ROC curve.

Results

Click to Enlarge

AnVil’s analysis provided the client with a succinct set of rules that they could use to accurately predict the toxicology of small molecules before they started the more costly process of synthesis and in vivo testing. AnVil further provided the client with an analysis of the impact of each structural feature and feature combination on predictive accuracy. Using the results of AnVil’s analysis, the client selected the set of structural features that would best minimize toxicology in accordance with its own specific utility measures. By using Anvil’s advanced data mining and visualization techniques to determine structure versus activity relationships instead of relying on experimental protocols, the client achieved significant cost savings.

To find out more about how AnVil can help you forge the optimal path to commercial discovery, send e-mail to info@anvilinfo.com.

back to top