Distributional semantic algorithms, such as word2vec, have truly revolutionized data science for text data. In a recently published pilot study, we have shown that such algorithms can also be used outside the realm of text, in the field of pharmacovigilance, and produce meaningful representations of important concepts such as drugs and side effects. Just like word vectors have tremendously improved performance of algorithms in various text tasks, such as machine translation, question-answering or text classification, we expect this new paradigm to provide a large boost in methods for the detection of previously unknow safety concerns and other pharmacovigilance methods.
- Reports of cases of side effects can be used like “sentences” with distributional semantic methods and provide meaningful new representations of drugs and side effects, which are traditionally represented simply as entries in terminologies
- The paradigm shift from terminologies to continuous representations allows for a much more nuanced definition of similarity, which will be very beneficial for advancing the science of pharmacovigilance
- Inspiration can come from anywhere, we have to stay curious, up-to-date and keep an eye on the large spectrum of data science applications