These are the codes and data mainly used for the project using deep tensor factorization for Predicting Anticander Drug Synergy. The only data set I used here is drug synergy data of 38 drugs and 39 cell lines, which is derived from the study of ONeil et al.
To implement the DTF model, firstly, I used R to preprocess the raw data to build the tensor to be used next for Python and MATLAB. For some specific cell lines, there were experiments carried out multiple times for the same drug pairs. In order to construct the three-dimensional (3D) drug-drug-cell-line tensor, we averaged these scores for the same drug-drug pairs. The resulting tensor is provided in the data_sets folder. To decompose the tensor in matlab, I employed the Tensor Toolbox and L-BFGS-B code. The results of the tensor decomposition are provided in the form of factor matrices. The python codes are mainly concerning how to generate features based on the output of Matlab and the codes to build the final DTF model to do predictions on missing drug combinations are given.
Note that for the convenience of programming, each drug combination was encoded into an index. To be specific, I used the formula: drug_A_index + drug_B_index 38 + cell_line_index 38 * 38, to encode drug combinations. Therefore, I can decode the index to get the original drug combination information. You can refer to the published paper for detailed algorithm. And if you have any questions, please feel free to contact me :).
If you would like to cite DTF, please cite the following publication:
Zexuan Sun, Shujun Huang, Peiran Jiang, Pingzhao Hu, DTF: Deep Tensor Factorization for predicting anticancer drug synergy, Bioinformatics, Volume 36, Issue 16, 15 August 2020, Pages 4483–4489, https://doi.org/10.1093/bioinformatics/btaa287.