According to a report by Harvard Medical School, as many as 90 percent of promising drug candidates fail before or during human clinical trials, falling into the so-called "valley of death". The Pharmaceutical Research and Manufacturers of America estimates that it takes an average of $2.6 billion and more than 10 years for a new medicine to hit the market.
Beforehand knowledge of pharmacokinetic properties like Absorption(A) coefficient, Distribution coefficient(D), Metabolism rate(M), Excretion rate(E), Toxicity(T) and Bioavailability for potential drugs in drug discovery trials can save a significant amount of time and money resources for researchers.
I developed different variations of Graph Convolutional Networks like Gated network, Attention network and, Gated-Augmented-Attention Networks(GAGCNs) for the task. The models were able to distinguish between drugs with higher and lower values of properties like Bioavailability with 88% accuracy.
The implementation code is propriety property of Aganitha Cognitive Solutions.