tests/testthat/test-running_melissa.R

In Melissa: Bayesian clustering and imputationa of single cell methylomes

context("Checking correct functionality of Melissa")

test_that("Melissa runs and returns correct results", {
  set.seed(1)
  obj <- melissa_synth_dt
  # Partition to train and test set
  expect_error(partition_dataset(obj, data_train_prcg = 1.4))
  expect_error(partition_dataset(obj, data_train_prcg = -1))
  expect_error(partition_dataset(obj, data_train_prcg = -1))
  expect_error(partition_dataset(obj, region_train_prcg = 1.4))
  expect_error(partition_dataset(obj, region_train_prcg = -1))

  # set.seed(1)
  # obj <- partition_dataset(obj, data_train_prcg = 0.5, region_train_prcg = 0.95,
  #                          cpg_train_prcg = 0.5, is_synth = FALSE)
  # expect_equal(NROW(obj$met[[1]][[1]]), 20)
  # expect_equal(NROW(obj$met[[1]][[12]]), 21)

  # # Run Melissa VB model
  # melissa_obj <- melissa(X = obj$met, K = 2, delta_0 = NULL, alpha_0 = 0.5,
  #                        vb_max_iter = 5, vb_init_nstart = 1,
  #                        is_parallel = FALSE, is_verbose = FALSE)
  #
  # # Check lots of parameters from Melissa output that match expectations
  # expect_gt(melissa_obj$W[1,1,1], 1.18)
  # expect_lt(melissa_obj$W[1,1,1], 1.19)
  #
  # expect_lt(melissa_obj$r_nk[1,2], 1e-20)
  #
  # expect_lt(melissa_obj$delta[1], 14)
  # expect_lt(melissa_obj$alpha[1], 61)
  # expect_lt(melissa_obj$beta[1], 39.19)
  # expect_lt(melissa_obj$lb[2], -4759.022)
  # expect_gt(melissa_obj$lb[2], -4759.9)

  # # Check imputation performance is correct
  # imputation_obj <- impute_met_state(obj = melissa_obj, test = obj$met_test)
  # expect_lt(imputation_obj$pred_obs[1], 0.515)
  # expect_gt(imputation_obj$pred_obs[1], 0.514)
  # expect_lt(melissa_obj$lb[2], -3426.022)
  #
  # # Check evaluation of imputation performance is correct
  # melissa_obj <- eval_imputation_performance(obj = melissa_obj,
  #                                            imputation_obj = imputation_obj)
  #
  # expect_lt(melissa_obj$imputation$auc, 0.9172)
  # expect_gt(melissa_obj$imputation$auc, 0.9171)
  #
  # expect_lt(melissa_obj$imputation$f_measure, 0.86812)
  # expect_gt(melissa_obj$imputation$f_measure, 0.86811)
  #
  # expect_lt(melissa_obj$imputation$pr@y.values[[1]][2], 0.85715)
  # expect_gt(melissa_obj$imputation$pr@y.values[[1]][2], 0.85714)


  # # Check the clustering performance is correct
  # melissa_obj <- eval_cluster_performance(melissa_obj, obj$opts$C_true)
  #
  # expect_lt(melissa_obj$clustering$ari, 1.00001)
  # expect_gt(melissa_obj$clustering$ari, 0.99999)
  #
  # expect_lt(melissa_obj$clustering$error, 0.001)
  # expect_gt(melissa_obj$clustering$error, -0.00001)


  # ##
  # # Test for Melissa Gibbs output
  # melissa_gibbs <- melissa_gibbs(X = obj$met, K = 2, gibbs_nsim = 10,
  #                                gibbs_burn_in = 5, is_parallel = FALSE,
  #                                is_verbose = FALSE)
  #
  # expect_lt(melissa_gibbs$summary$pi[1], 0.668493)
  # expect_gt(melissa_gibbs$summary$pi[1], 0.668492)
  #
  # expect_lt(melissa_gibbs$r_nk[2,1], 1.001)
  # expect_gt(melissa_gibbs$r_nk[2,1], 0.999)
  #
  # expect_lt(melissa_gibbs$W[1,1,1], 1.3791)
  # expect_gt(melissa_gibbs$W[1,1,1], 1.3790)
})