a transfer-learning based Cox proportional hazards network (TCAP) by utilizing an integrated loss function consisted of two modules: the data reconstruction module to ensure learning a representative hidden layer for the input data, and the proportional hazard module to estimate patients’ risks.
tensorflow, python 3.7, lifelines
In this study, we utilized cancer datasets from the TCGA portal (https://tcga-data.nci.nih.gov/tcga/). All these datasets were downloaded by using the R package “TCGA-assembler”(v1.0.3, (Wei, et al., 2018)), which contains four types of multi-omics data: mRNA, miRNA, DNA methylation, and copy number variation (CNV) data. Here, “mRNA” was RNA sequencing data generated by UNC Illumina HiSeq_RNASeq V2; Level 3, “miRNA” was miRNA sequencing data obtained by BCGSC Illumina HiSeq miRNASeq, DNA methylation data was generated by USC HumanMethylation450, and CNV data that generated by BROAD-MIT Genome wide SNP_6.
DCAP was a framework with three different method. Before using you should choose the training data in dataset.py, the test.py is used for choosing the hyper-parameters by k-fold CV and allpred.py is used to construct the prediction model by using all data.
For easy to use, here we give four example data: brca1.csv used in test.py or allpred.py
This method is still on progress, any questions can be sent to [email protected]