This repository contains R implementation of the algorithms proposed in "A multitask multiple kernel learning formulation for discriminating early- and late-stage cancers" (Bioinformatics).

• run_mtmkl_step1_coclustering.R : shows how to produce similarity matrices based on the TCGA cohorts and Hallmark pathways
• run_mtmkl_step2_collect_coclustering_data.R : produces the similarity matrices by combining the results obtained in step 1
• run_mtmkl_step3_using_aggregated_similarity_matrices.R : shows how to replicate multitask experiments on the TCGA cohorts using the similarity matrices produced in step 2.
• run_mtmkl_step4_collect_classification_data.R : collects the classification results

MTGSBC methods

• classification_helper.R => helper functions
• solve_classification_models_cplex.R => support vector machine classification solver, and the cutting plane model using CPLEX optimization software
• solve_classification_models_mosek.R => support vector machine classification solver, and the cutting plane model using Mosek optimization software
• group_lasso_multitask_multiple_kernel_classification_train.R => training procedure for multitask group Lasso MKL
• group_lasso_multitask_multiple_kernel_classification_test.R => test procedure for multitask group Lasso MKL
• mtmkl_coclustering_algorithm.R => the heuristic algorithm used for the diversificaion phase in step 1