This is the code that accompanies our paper BenMO: Benchmarking predictive machine learning algorithms using MultiOmics data. In this repository, we provide R code for data pre-processing and a combination of Python and R code for modeling, results, and plots. We also provide Python wrappers for the models that are run using R functions. In addition, we provide model wrappers for optimizing the hyperparameters of the models that do not have the corresponding function currently implemented in Python or R.
The healthy pregnancy dataset contains only samples from women who delivered at term with no complications and is available from http://nalab.stanford.edu/multiomics-pregnancy/. The preeclampsia dataset contains both healthy and preeclamptic pregnancies, and is available from https://github.com/ivanam5/Multiomics_Preeclampsia/tree/main/Multiomics_Datasets. The TCGA data we use here are related to 10 types of cancer, Breast invasive carcinoma (TCGA-BRCA), Colorectal adenocarcinoma (TCGA-COADREAD), Glioma (TCGA-GBMLGG), Head and Neck squamous cell carcinoma (TCGA-HNSC), Pan-kidney cohort (TCGA-KIPAN), Lung squamous cell carcinoma (TCGA-LUSC), Ovarian serous cystadenocarcinoma (TCGA-OV), Stomach and Esophageal carcinoma (TCGA-STES), Thyroid carcinoma (TCGA-THCA), and Uterine corpus endometrial carcinoma (TCGA-UCEC). Datasets are available from http://linkedomics.org/. The majority of TCGA studies contain 5 modalities - miRNA (Hi Seq, Gene level), RNAseq (Hi Seq, Gene level), RPPA (Reverse Phase Protein Array) (Gene level), SCNV (Somatic Copy Number Variation) (Gene level, log-ratio), and DNA methylation (Gene level, HM450K), with the exception of uterine cancer that excludes the RNA modality.