|Image from Paper that Shows Training of DNN for Drug Discovery|
Monday, May 30, 2016
Yet another exciting paper published by InSilico Inc in addition to their work on developing the model based on Deep Neural Networks on the prediction of human age... (paper can be downloaded here). Summary of the work can be seen in our previous blog posts here
This month paper was published about the use of Deep Neural Networks (DNN) in new drug discovery and repurposing using transcriptomic data.
The paper got recently published in Molecular Pharmaceutics....... can be downloaded from here
Deep learning is an artificial intelligence (AI). It utilises higher level or multilayer of the neuron to model the high level of abstraction of data.
The paper published clearly shows that AI neural network was able to predict the therapeutic use of a large number of medicine depending on gene expression data that is obtained from high-throughput experiments on human cell lines.
The significance of the work got recently published in Eureka News Alert
The study uses 678 drugs affecting A549, MCF-7 and PC-3 cell lines from LINCS library developed by NIH that is linked to 12 therapeutic categories established by MeSH (Medical Subject Heading). The library is maintained by NLM. So basically researchers trained DNN by utilising both transcriptomic data using a scoring algorithm for samples that are perturbed with different concentration of the drug after 6 or 24 hours. The cross-validation of the model showed that DNN achieved 54.6% accuracy in correctly predicting one out of 12 therapeutic categories for each drug. Interestingly, a large number of the drugs that were misclassified by DNN was found to have dual therapeutic utility. This suggests that may be the confusion matrix used for DNN can be used for drug repurposing
This is the first study where DNN based model was developed on the basis of transcriptomic data for predicting the therapeutic use of the drug. Hence, it was proof of concept study that DNNs can be used for annotation of drugs using transcriptomic signatures. Hence, they used this finding to progress towards the development of a pipeline program in order to accelerate the preclinical of drugs for most any therapeutic category. They believe that if this technique can be extrapolated to invite signatures then maybe this can double the number of molecules in drug discovery studies.