2016
cMapper allows biologists to interrogate data points in the EBI-RDF platform that are connected to genes or small molecules of interest in multiple biological contexts. Input to cMapper consists of a set of genes or small molecules, and the output is the data points in six independent EBI-RDF databases connected with the given genes or small molecules in the user's query. cMapper pro-vides output to users in the form of a graph in which nodes represent data points and the edges represent connections between data points and inputted sets of genes or small molecules. Users can also apply filters based on database, taxonomy, organ and pathways in order to focus on a core connectivity graph of their interests.
Deep Learning for Drug Prediction using Gene Expression Data
2018
In this project, we designed a deep learning framework to predict drug response using gene expression data. Our Drug Prediction framework consists of following three different stages (1) pre-processing, (2) feature selection and (3) model fitting. The pre-processing phase gene expression values were normalized to avoid from biasness using min-max transformation which alighted the data points between 0 and 1. Since it was not clear to us that which features selection method would perform better therefore we performed an evaluation test using Principle Component Analysis, Mutual Information, Differential Expression Analysis, and Chi Squires test. We designed and tested deep neural networks with multiple architectures in term of number of hidden layers and number of neurons in each layer. We perform experiments for 2, 4, 8, 16 hidden layers. DNN was designed using Keras (A python Library) and performance was measured using scikit learn The performance of DNN archticutres was analyzed based on Learning Rate and Accracy. The developed deep neural network to predicted drug response of five cancer drugs with more 95% accuracy and predict drug targets of 9 drugs with more than 90% accuracy using the gene expression data.