Predicting key example compounds in competitors' patent applications using structural information alone

In drug discovery programs, predicting key example compounds in competitors' patent applications is important work for scientists working in the same or in related research areas. In general, medicinal chemists are responsible for this work, and they attempt to guess the identity of key compounds based on information provided in patent applications, such as biological data, scale of reaction, and/or optimization of the salt form for a particular compound. However, this is sometimes made difficult by the lack of such information. This paper describes a method for predicting key compounds in competitors' patent applications by using only structural information of example compounds. Based on the assumption that medicinal chemists usually carry out extensive structure-activity relationship (SAR) studies around key compounds, the method identifies compounds located at the centers of densely populated regions in the patent examples' chemical space, as represented by Extended Connectivity Fingerprints (ECFPs). For the validation of the method, a total of 30 patents containing structures of launched drugs were selected to test whether or not the method is able to predict key compounds (the launched drugs). In 17 out of the 30 patents (57%), the method was able to successfully predict the key compounds. The result indicates that our method could provide an alternative approach to predicting key compounds in cases where the conventional medicinal chemist's approach does not work well. This method could also be used as a complement to the traditional medicinal chemist's approach.