Datasets
Models
Diff of
/R/job_grid.R
[000000]
..
[c3b4f8]
Switch to side-by-side view
--- a +++ b/R/job_grid.R @@ -0,0 +1,731 @@ +# ${TRAIN_FILE}${GPU_PER_TRIAL}${NUM_SAMPLES}${N_FOLDS}${DATA_TYPES}${NAME_TAG}${SUBSET_TYPE}${STRATIFY}${BOTTLENECK}${FULL}${ENCODER_TRAIN} +require(utils) +require(data.table) +all_combos <- c( + # "gnndrug mut exp", + # "gnndrug cnv exp", + # "gnndrug exp prot", + # "gnndrug exp rppa", + # "gnndrug exp hist", + # "gnndrug exp metab", + # "gnndrug exp mirna", + # "gnndrug prot rppa", + # "gnndrug cnv prot", + "gnndrug mut cnv", + # "gnndrug mirna metab", + # "gnndrug metab hist", + # "gnndrug metab rppa", + + # "drug cnv exp metab" + + # "gnndrug cnv exp prot", + # "gnndrug cnv exp prot mirna metab hist rppa", + # "gnndrug exp rppa hist prot", + # "gnndrug exp prot hist rppa", + # "gnndrug exp prot rppa", + # "gnndrug exp rppa prot", + # "gnndrug mirna metab hist rppa", + # "gnndrug mut cnv exp prot" + # "gnndrug mut cnv exp prot mirna metab hist rppa" + + # "gnndrug mut cnv exp", + + # "gnndrug cnv exp prot metab" + # "drug cnv exp prot metab" + + # "drug mut cnv exp", + # "drug mut cnv exp prot", + # "drug cnv exp prot", + # "drug exp rppa prot", + # "drug exp rppa hist prot", + # "drug mirna metab hist rppa", + # "drug mut cnv exp prot mirna metab hist rppa", + # "drug cnv exp prot mirna metab hist rppa" + + # "gnndrug mut prot", + # "gnndrug mut mirna", + # "gnndrug mut metab", + # "gnndrug mut hist", + # "gnndrug mut rppa", + # "gnndrug cnv mirna", + # "gnndrug cnv metab", + # "gnndrug cnv hist", + # "gnndrug cnv rppa", + # "gnndrug prot mirna", + # "gnndrug prot metab", + # "gnndrug prot hist", + "gnndrug mirna hist", + "gnndrug mirna rppa", + "gnndrug hist rppa", + + # "drug mut prot", + # "drug mut mirna", + # "drug mut metab", + # "drug mut hist", + # "drug mut rppa", + # "drug cnv mirna", + "drug cnv metab", + # "drug cnv hist", + # "drug cnv rppa", + # "drug prot mirna", + # "drug prot metab", + # "drug prot hist", + # "drug mirna hist", + # "drug mirna rppa", + # "drug hist rppa", + # + # "drug mut exp", + # "drug cnv exp", + # "drug exp prot", + # "drug exp rppa", + # "drug exp hist", + # "drug exp metab", + # "drug exp mirna", + # "drug prot rppa", + # "drug cnv prot", + # "drug mut cnv", + # "drug mirna metab", + # "drug metab hist", + # "drug metab rppa" + + + # "gnndrug exp", + "gnndrug mut", + "gnndrug cnv", + "gnndrug prot", + "gnndrug mirna" + # "gnndrug metab", + # "gnndrug hist", + # "gnndrug rppa" + # + # "drug exp", + # "drug mut", + # "drug cnv", + # "drug prot", + # "drug mirna", + # "drug metab" + # "drug hist", + # "drug rppa" + + # "gnndrug mut exp hist", + # "gnndrug cnv exp hist", + # "gnndrug exp prot hist", + # "gnndrug exp mirna hist", + # "gnndrug exp metab hist", + # "gnndrug exp hist rppa" + # "gnndrug cnv exp metab" + + ) +# ${TRAIN_FILE} ${N_FOLDS} ${DATA_TYPES} ${NAME_TAG} ${SUBSET_TYPE} ${STRATIFY} ${FULL} ${ENCODER_TRAIN} + +all_grids <- vector(mode = "list") +for (combo in all_combos) { + ENCODER_TRAIN <- "1" + NUM_SAMPLES <- "40" + if (grepl("cnv", combo)) { + GPU_PER_TRIAL <- "1" + } else if (combo == "gnndrug prot" | combo == "drug prot") { + GPU_PER_TRIAL <- "0.5" + # NUM_SAMPLES <- "32" + } else if (combo == "gnndrug exp" | + combo == "gnndrug exp prot" | + combo == "gnndrug mut" | + combo == "drug mut" | + combo == "drug exp" | combo == "drug exp prot") { + GPU_PER_TRIAL <- "0.5" + # NUM_SAMPLES <- "8" + } else if (combo == "gnndrug mirna" | + combo == "gnndrug metab" | + combo == "gnndrug hist" | combo == "gnndrug rppa" | + combo == "drug mirna" | + combo == "drug metab" | + combo == "drug hist" | combo == "drug rppa") { + GPU_PER_TRIAL <- "0.2" + # NUM_SAMPLES <- "40" + } else { + GPU_PER_TRIAL <- "1" + # NUM_SAMPLES <- "8" + } + + # LOSS_TYPE = "rmse" + # loss_type_name = "RMSELoss" + # LOSS_TYPE = "weighted_rmse" + # loss_type_name = "WeightedRMSELoss" + + for (LOSS_TYPE in c("rmse", "weighted_rmse")) { + if (LOSS_TYPE == "rmse") { + loss_type_name = "RMSELoss" + } else { + loss_type_name = "WeightedRMSELoss" + } + for (ONE_HOT_DRUGS in c("0", "1")) { + if (ONE_HOT_DRUGS == "1") { + one_hot_drugs_name = "OneHotDrugs" + } else { + if (grepl("gnndrug", combo) == TRUE) { + one_hot_drugs_name = "GNNDrugs" + } else { + one_hot_drugs_name = "MorganDrugs" + } + } + for (MERGE_METHOD in c("sum", "concat", "lmf")) { + if (MERGE_METHOD == "sum") { + merge_method_name = "MergeBySum" + } else if (MERGE_METHOD == "concat") { + merge_method_name = "MergeByConcat" + } else if (MERGE_METHOD == "lmf") { + merge_method_name = "MergeByLMF" + } + for (BOTTLENECK in c("0", "1")) { + # for (TRAIN_FILE in c("CTRP_AAC_MORGAN_1024.hdf")) { + # for (TRAIN_FILE in c("CTRP_AAC_SMILES.txt", "GDSC1_AAC_SMILES.txt", "GDSC2_AAC_SMILES.txt")) { + # for (TRAIN_FILE in c("CTRP_AAC_MORGAN_1024.hdf", "GDSC1_AAC_MORGAN_1024.hdf", "GDSC2_AAC_MORGAN_1024.hdf")) { + # for (TRAIN_FILE in c("CTRP_AAC_SMILES.txt")) { + if (grepl("gnn", combo)) { + TRAIN_FILE = "CTRP_AAC_SMILES.txt" + } else { + TRAIN_FILE = "CTRP_AAC_MORGAN_1024.hdf" + } + train_set_name <- "CTRP" + for (SUBSET_TYPE in c("cell_line", "drug", "lineage", "both")) { + if (SUBSET_TYPE == "both") { + N_FOLDS <- "5" + } else { + N_FOLDS <- "5" + } + for (FULL in c("0", "1")) { + if (FULL == "1") { + cur_pretrain <- "0" + # train_set_name <- gsub("\\_.+", "", TRAIN_FILE) + full <- "FullModel" + encoder <- "EncoderTrain" + split <- toupper(SUBSET_TYPE) + data_types <- gsub(" ", "_", combo) + if (BOTTLENECK == "0") { + bottleneck <- "NoBottleNeck" + } else { + bottleneck <- "WithBottleNeck" + } + if (cur_pretrain == "0") { + pretrain <- "NoTCGAPretrain" + } else { + pretrain <- "WithTCGAPretrain" + } + + NAME_TAG <- + paste( + "HyperOpt_DRP", + train_set_name, + full, + encoder, + "Split", + split, + bottleneck, + pretrain, + merge_method_name, + loss_type_name, + one_hot_drugs_name, + data_types, + sep = "_" + ) + + cur_grid <- data.table( + TRAIN_FILE = TRAIN_FILE, + GPU_PER_TRIAL = GPU_PER_TRIAL, + NUM_SAMPLES = NUM_SAMPLES, + N_FOLDS = N_FOLDS, + DATA_TYPES = combo, + NAME_TAG = NAME_TAG, + SUBSET_TYPE = SUBSET_TYPE, + STRATIFY = "1", + BOTTLENECK = BOTTLENECK, + FULL = FULL, + ENCODER_TRAIN = ENCODER_TRAIN, + PRETRAIN = cur_pretrain, + MERGE_METHOD = MERGE_METHOD, + LOSS_TYPE = LOSS_TYPE, + ONE_HOT_DRUGS = ONE_HOT_DRUGS + ) + all_grids <- append(all_grids, list(cur_grid)) + + } else { + if (grepl("cnv", combo) | grepl("exp", combo)) { + for (PRETRAIN in c("0", "1")) { + cur_encoder_train <- "1" + encoder <- "EncoderTrain" + # train_set_name <- gsub("\\_.+", "", TRAIN_FILE) + full <- "ResponseOnly" + split <- toupper(SUBSET_TYPE) + data_types <- gsub(" ", "_", combo) + if (BOTTLENECK == "0") { + bottleneck <- "NoBottleNeck" + } else { + bottleneck <- "WithBottleNeck" + } + if (PRETRAIN == "0") { + pretrain <- "NoTCGAPretrain" + } else { + pretrain <- "WithTCGAPretrain" + } + + NAME_TAG <- + paste( + "HyperOpt_DRP", + train_set_name, + full, + encoder, + "Split", + split, + bottleneck, + pretrain, + merge_method_name, + loss_type_name, + one_hot_drugs_name, + data_types, + sep = "_" + ) + + cur_grid <- data.table( + TRAIN_FILE = TRAIN_FILE, + GPU_PER_TRIAL = GPU_PER_TRIAL, + NUM_SAMPLES = NUM_SAMPLES, + N_FOLDS = N_FOLDS, + DATA_TYPES = combo, + NAME_TAG = NAME_TAG, + SUBSET_TYPE = SUBSET_TYPE, + STRATIFY = "1", + BOTTLENECK = BOTTLENECK, + FULL = FULL, + ENCODER_TRAIN = cur_encoder_train, + PRETRAIN = PRETRAIN, + MERGE_METHOD = MERGE_METHOD, + LOSS_TYPE = LOSS_TYPE, + ONE_HOT_DRUGS = ONE_HOT_DRUGS + ) + all_grids <- append(all_grids, list(cur_grid)) + + } + } else { + cur_encoder_train <- "1" + encoder <- "EncoderTrain" + # train_set_name <- gsub("\\_.+", "", TRAIN_FILE) + full <- "ResponseOnly" + split <- toupper(SUBSET_TYPE) + data_types <- gsub(" ", "_", combo) + if (BOTTLENECK == "0") { + bottleneck <- "NoBottleNeck" + } else { + bottleneck <- "WithBottleNeck" + } + cur_pretrain <- "0" + pretrain <- "NoTCGAPretrain" + + NAME_TAG <- + paste( + "HyperOpt_DRP", + train_set_name, + full, + encoder, + "Split", + split, + bottleneck, + pretrain, + merge_method_name, + loss_type_name, + one_hot_drugs_name, + data_types, + sep = "_" + ) + + cur_grid <- data.table( + TRAIN_FILE = TRAIN_FILE, + GPU_PER_TRIAL = GPU_PER_TRIAL, + NUM_SAMPLES = NUM_SAMPLES, + N_FOLDS = N_FOLDS, + DATA_TYPES = combo, + NAME_TAG = NAME_TAG, + SUBSET_TYPE = SUBSET_TYPE, + STRATIFY = "1", + BOTTLENECK = BOTTLENECK, + FULL = FULL, + ENCODER_TRAIN = cur_encoder_train, + PRETRAIN = cur_pretrain, + MERGE_METHOD = MERGE_METHOD, + LOSS_TYPE = LOSS_TYPE, + ONE_HOT_DRUGS = ONE_HOT_DRUGS + ) + all_grids <- append(all_grids, list(cur_grid)) + } + } + } + } + } + } + } + } +} +all_param_combos <- rbindlist(all_grids) + + +# ==== Bi-Modal split by DRUG (ALL) ==== +combos <- all_param_combos[PRETRAIN == 0 & SUBSET_TYPE == "drug" & MERGE_METHOD == "concat" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +# combos <- all_param_combos[PRETRAIN == 0 & SUBSET_TYPE == "drug" & MERGE_METHOD == "lmf" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- all_param_combos[PRETRAIN == 0 & SUBSET_TYPE == "drug" & MERGE_METHOD == "concat" & LOSS_TYPE == "rmse" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +# fwrite(combos, "DRP/slurm/grids/drp_cv_split_by_DRUG_grid.csv", col.names = F) +# fwrite(combos, "DRP/slurm/grids/drp_cv_lmf_split_by_DRUG_grid.csv", col.names = F) +# fwrite(combos, "DRP/slurm/grids/drp_validate_lmf_split_by_DRUG_grid.csv", col.names = F) +fwrite(combos, "DRP/slurm/grids/drp_validate_split_by_DRUG_grid.csv", col.names = F) + +# ==== Bi-Modal split by CELL_LINE (ALL) ==== +combos <- all_param_combos[PRETRAIN == 0 & SUBSET_TYPE == "cell_line" & (MERGE_METHOD == "concat" | MERGE_METHOD == "lmf") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_cv_split_by_CELL_LINE_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & SUBSET_TYPE == "cell_line" & (MERGE_METHOD == "concat" | MERGE_METHOD == "lmf") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +# fwrite(combos, "DRP/slurm/grids/drp_validate_split_by_CELL_LINE_grid.csv", col.names = F) + +# ==== (GDSC) Bi-modal with GNN + LMF ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/drp_opt_gdsc_grid.csv", col.names = F) + +# ==== Bi-modal Baseline (Morgan + Concat + RMSE) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_baseline_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & (SUBSET_TYPE == "cell_line" | SUBSET_TYPE == "drug") & LOSS_TYPE == "rmse" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_baseline_cell_line_drug_grid.csv", col.names = F) + +# ==== Bi-modal Baseline + LDS (Morgan + Concat + WeightedRMSE) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_baseline_grid.csv", col.names = F) +# combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & (SUBSET_TYPE == "cell_line" | SUBSET_TYPE == "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_baseline_with_lds_cell_line_drug_grid.csv", col.names = F) + +# ==== Bi-modal Baseline + LMF (Morgan + LMF + RMSE) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0 & LOSS_TYPE == "rmse"] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_baseline_with_lmf_grid.csv", col.names = F) +# combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & (SUBSET_TYPE == "cell_line" | SUBSET_TYPE == "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & SUBSET_TYPE != "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0 & LOSS_TYPE == "rmse"] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_baseline_with_lmf_cell_line_drug_grid.csv", col.names = F) + +# ==== Bi-modal Baseline + LMF (Morgan + LMF + RMSE) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "sum") & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0 & LOSS_TYPE == "rmse"] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_baseline_with_sum_grid.csv", col.names = F) +# combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & (SUBSET_TYPE == "cell_line" | SUBSET_TYPE == "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "sum") & SUBSET_TYPE != "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0 & LOSS_TYPE == "rmse"] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_baseline_with_sum_cell_line_drug_grid.csv", col.names = F) + +# ==== Bi-modal Baseline + GNN (GNN + concat + RMSE) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0 & LOSS_TYPE == "rmse"] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_baseline_with_gnn_grid.csv", col.names = F) +# combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & (SUBSET_TYPE == "cell_line" | SUBSET_TYPE == "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & SUBSET_TYPE != "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0 & LOSS_TYPE == "rmse"] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_baseline_with_gnn_cell_line_drug_grid.csv", col.names = F) +# fwrite(combos, "DRP/slurm/grids/drp_opt_baseline_with_gnn_cell_line_drug_grid_extra.csv", col.names = F) + + +# ==== Bi-modal with GNN + LMF but no LDS ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/drp_opt_noLDS_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & (SUBSET_TYPE == "cell_line" | SUBSET_TYPE == "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/drp_opt_noLDS_cell_line_drug_grid.csv", col.names = F) + +# ==== Bi-modal GNN + Concatenation (non-LMF) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "weighted_rmse" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_concat_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & (SUBSET_TYPE == "cell_line" | SUBSET_TYPE == "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_concat_cell_line_drug_grid.csv", col.names = F) +# combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & SUBSET_TYPE == "drug" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +# combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +# fwrite(combos, "DRP/slurm/drp_opt_concat_drug_grid.csv", col.names = F) + +# ==== Bi-modal GNN + Sum (non-LMF) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "sum") & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/drp_opt_sum_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "sum") & (SUBSET_TYPE == "cell_line" | SUBSET_TYPE == "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_sum_cell_line_drug_grid.csv", col.names = F) + +# ==== Bi-modal GNN + LMF + LDS (trifecta) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_bi_lmf_lds_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & (SUBSET_TYPE == "cell_line" | SUBSET_TYPE == "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +# combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_bi_lmf_lds_cell_line_drug_grid.csv", col.names = F) + +# ==== Bi-modal GNN + LMF without LDS (trifecta - LDS) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "rmse" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_bi_trifecta_without_lds_grid.csv", col.names = F) +# combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & (SUBSET_TYPE == "cell_line" | SUBSET_TYPE == "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] + +# ==== Bi-modal LDS + GNN without LMF (trifecta - LMF) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & LOSS_TYPE == "weighted_rmse" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_bi_trifecta_without_lds_grid.csv", col.names = F) +# combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & (SUBSET_TYPE == "cell_line" | SUBSET_TYPE == "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] + +# ==== Bi-modal LMF + LDS without GNN ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & + SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_morgan_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & (SUBSET_TYPE == "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_morgan_cell_line_drug_grid.csv", col.names = F) + +# ==== Bi-modal Trifecta ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_bimodal_trifecta_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & SUBSET_TYPE == "drug" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_bimodal_trifecta_cell_line_drug_grid.csv", col.names = F) + +# ==== Bi-Modal Bottleneck (Cell Line Splitting/Grouping) ==== +baseline_combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "rmse" & + TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & SUBSET_TYPE == "cell_line" & + BOTTLENECK == 1 & FULL == 1 & ONE_HOT_DRUGS == 0] +baseline_combos <- baseline_combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(baseline_combos, "DRP/slurm/grids/drp_opt_bimodal_baseline_bottleneck_cell_line_grid.csv", col.names = F) + +trifecta_combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & + TRAIN_FILE == "CTRP_AAC_SMILES.txt" & SUBSET_TYPE == "cell_line" & + BOTTLENECK == 1 & FULL == 1 & ONE_HOT_DRUGS == 0] +trifecta_combos <- trifecta_combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(trifecta_combos, "DRP/slurm/grids/drp_opt_bimodal_trifecta_bottleneck_cell_line_grid.csv", col.names = F) + + +# ==== Tri-modal Trifecta (initial) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_trimodal_trifecta_other_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & DATA_TYPES %like% "gnn" & LOSS_TYPE == "weighted_rmse" & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & SUBSET_TYPE == "cell_line" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_trimodal_trifecta_other_cell_line_drug_grid.csv", col.names = F) + +# ==== Tri-modal Trifecta (remainders) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_trimodal_trifecta_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & (SUBSET_TYPE == "cell_line" | SUBSET_TYPE == "cell_line") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_trimodal_trifecta_cell_line_drug_grid.csv", col.names = F) + +# ==== Tri-modal Trifecta, split by lineage ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & SUBSET_TYPE == "lineage" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_trimodal_trifecta_lineage_grid.csv", col.names = F) + +# ==== Tri-modal baseline (remainders) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "rmse" & TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_trimodal_baseline_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "rmse" & TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & SUBSET_TYPE != "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_trimodal_baseline_cell_line_drug_grid.csv", col.names = F) + +# ==== Tri-modal baseline (remainders) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "rmse" & TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_trimodal_baseline_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "rmse" & TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & SUBSET_TYPE != "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_trimodal_baseline_cell_line_drug_grid.csv", col.names = F) + +# ==== Tri-Modal Bottleneck (Cell Line Splitting/Grouping) ==== +baseline_combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "rmse" & + TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & SUBSET_TYPE == "cell_line" & + BOTTLENECK == 1 & FULL == 1 & ONE_HOT_DRUGS == 0] +baseline_combos <- baseline_combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(baseline_combos, "DRP/slurm/grids/drp_opt_trimodal_baseline_bottleneck_cell_line_grid.csv", col.names = F) + +trifecta_combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & + TRAIN_FILE == "CTRP_AAC_SMILES.txt" & SUBSET_TYPE == "cell_line" & + BOTTLENECK == 1 & FULL == 1 & ONE_HOT_DRUGS == 0] +trifecta_combos <- trifecta_combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(trifecta_combos, "DRP/slurm/grids/drp_opt_trimodal_trifecta_bottleneck_cell_line_grid.csv", col.names = F) + +# EXP_HIST +# MIRNA_RPPA +# ==== Multi-modal Trifecta with EXP_HIST ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_exp_hist_trifecta_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & SUBSET_TYPE != "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_exp_hist_trifecta_cell_line_drug_grid.csv", col.names = F) +# ==== Multi-modal Trifecta with CNV_EXP_METAB ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +combos$GPU_PER_TRIAL <- 0.5 +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_cnv_exp_metab_trifecta_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & SUBSET_TYPE != "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +combos$GPU_PER_TRIAL <- 0.5 +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_cnv_exp_metab_trifecta_cell_line_drug_grid.csv", col.names = F) + +# ==== Multi-modal Trifecta with CNV_EXP_PROT_METAB ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +combos$GPU_PER_TRIAL <- 1 +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_cnv_exp_metab_trifecta_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +combos$GPU_PER_TRIAL <- 1 +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_cnv_exp_prot_metab_trifecta_cell_line_drug_grid.csv", col.names = F) + +# ==== Multi-modal Baseline ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "rmse" & TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_baseline_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "rmse" & TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & SUBSET_TYPE != "cell_line" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_baseline_cell_line_drug_grid.csv", col.names = F) + +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "rmse" & TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_baseline_all_grid.csv", col.names = F) + +# ==== Multi-modal Baseline + LMF ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "rmse" & TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_baseline_with_lmf_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "rmse" & TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & SUBSET_TYPE != "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_baseline_with_lmf_cell_line_drug_grid.csv", col.names = F) + +# ==== Multi-modal Baseline + LDS ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "weighted_rmse" & TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_baseline_with_lds_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "rmse" & TRAIN_FILE == "CTRP_AAC_MORGAN_1024.hdf" & SUBSET_TYPE != "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_baseline_with_lmf_cell_line_drug_grid.csv", col.names = F) + +# ==== Multi-modal GNN + LMF + LDS (trifecta) ==== +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_trifecta_grid.csv", col.names = F) +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & TRAIN_FILE == "CTRP_AAC_SMILES.txt" & SUBSET_TYPE != "lineage" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_multimodal_trifecta_cell_line_drug_grid.csv", col.names = F) + +table(all_param_combos$DATA_TYPES) +all_param_combos[PRETRAIN == 0] +all_param_combos[PRETRAIN == 1] +unique(all_param_combos[, .SD, .SDcols = !c("PRETRAIN")]) +all_param_combos[PRETRAIN == 0 & FULL == 1] +all_param_combos[PRETRAIN == 0 & FULL == 1] +all_param_combos[FULL == 1] +table(all_param_combos[FULL == 1]$DATA_TYPES) +table(all_param_combos[FULL == 1 & PRETRAIN == 0]$DATA_TYPES) +all_param_combos[FULL == 1 & PRETRAIN == 0 & MERGE_METHOD == "concat" & SUBSET_TYPE == "drug"] +all_param_combos[FULL == 0 & PRETRAIN == 0 & MERGE_METHOD == "concat" & SUBSET_TYPE == "drug"] + +# 2 data types including exp +combos <- all_param_combos[PRETRAIN == 0 & MERGE_METHOD == "sum" & SUBSET_TYPE == "drug" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +fwrite(combos, "DRP/slurm/drp_opt_grid.csv", col.names = F) +# unimodal + bimodal with exp + lmf + gnndrug +combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +combos <- combos[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/drp_opt_grid.csv", col.names = F) +# fwrite(combos, "DRP/slurm/drp_opt_grid_sub.csv", col.names = F) + +combos_1 <- all_param_combos[PRETRAIN == 0 & MERGE_METHOD == "sum" & SUBSET_TYPE == "drug" & BOTTLENECK == 0 & FULL == 1] +combos_2 <- all_param_combos[PRETRAIN == 0 & MERGE_METHOD == "sum" & SUBSET_TYPE == "drug" & BOTTLENECK == 0 & FULL == 0 & DATA_TYPES == "drug cnv"] +combos_3 <- all_param_combos[PRETRAIN == 0 & MERGE_METHOD == "sum" & SUBSET_TYPE == "drug" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 1] +fwrite(unique(rbindlist(list(combos_1, combos_2, combos_3))), "DRP/slurm/drp_opt_grid.csv", col.names = F) + + +fwrite(all_param_combos[PRETRAIN == 0 & MERGE_METHOD == "sum" & SUBSET_TYPE == "drug" & BOTTLENECK == 0], "DRP/slurm/drp_opt_grid.csv", col.names = F) +fwrite(all_param_combos[PRETRAIN == 0 & MERGE_METHOD == "sum" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & ONE_HOT_DRUGS == 0], "DRP/slurm/drp_opt_grid.csv", col.names = F) +fwrite(all_param_combos[PRETRAIN == 0 & FULL == 0 & MERGE_METHOD == "concat" & SUBSET_TYPE == "drug" & ONE_HOT_DRUGS == 0], "DRP/slurm/drp_opt_grid.csv", col.names = F) + +fwrite(all_param_combos[PRETRAIN == 0 & FULL == 1], "DRP/slurm/drp_opt_extra_grid.csv", col.names = F) +fwrite(all_param_combos[PRETRAIN == 0 & FULL == 1][1], "DRP/slurm/drp_opt_test_grid.csv", col.names = F) +fwrite(all_param_combos[PRETRAIN == 0 & FULL == 1], "DRP/slurm/drp_opt_drug_grid.csv", col.names = F) + +colnames(all_param_combos) + +# "TRAIN_FILE" "GPU_PER_TRIAL" "NUM_SAMPLES" "N_FOLDS" "DATA_TYPES" "NAME_TAG" "SUBSET_TYPE" "STRATIFY" "BOTTLENECK" +# "FULL" "ENCODER_TRAIN" "PRETRAIN" "MERGE_METHOD" "LOSS_TYPE" "ONE_HOT_DRUGS" +# ==== CV Grid ==== +# require(data.table) +# cur_opt_grid <- fread("DRP/slurm/drp_opt_grid.csv") +# # ${TRAIN_FILE} ${N_FOLDS} ${DATA_TYPES} ${NAME_TAG} ${SUBSET_TYPE} ${STRATIFY} ${FULL} ${ENCODER_TRAIN} +# cur_cv_grid <- unique(cur_opt_grid[, c(1, 4, 5, 6, 7, 8, 10, 11, 13, 14, 15)]) +# fwrite(cur_cv_grid, "DRP/slurm/drp_cv_grid.csv", col.names = F) +# +# cur_cv_grid[V6 %like% ".*ResponseOnly.*"] +# fwrite(cur_cv_grid[V6 %like% ".*ResponseOnly.*" & V5 != "drug cnv" & V15 == 0], "DRP/slurm/drp_cv_grid.csv", col.names = F) +# fwrite(cur_cv_grid[!(V6 %like% ".*ResponseOnly.*" & V5 == "drug cnv")], "DRP/slurm/drp_cv_grid.csv", col.names = F) +# fwrite(cur_cv_grid[!(V6 %like% ".*ResponseOnly.*" & V5 == "drug cnv") & !(V6 %like% "OneHotDrugs")], "DRP/slurm/drp_infer_grid.csv", col.names = F) +# +# cur_cv_grid <- fread("DRP/slurm/drp_cv_grid.csv") +# cur_cv_grid$V1 <- gsub(pattern = "CTRP", replacement = "GDSC2", cur_cv_grid$V1) +# cur_cv_grid$V4 <- gsub(pattern = "CTRP", replacement = "GDSC2", cur_cv_grid$V4) +# +# +# +# fwrite(cur_cv_grid, "DRP/slurm/drp_cv_grid.csv", col.names = F) + +# Remaining CV Runs ==== +cnv_metab_base <- all_param_combos[DATA_TYPES == "drug cnv metab" & PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "rmse" & + SUBSET_TYPE == "cell_line" & BOTTLENECK == 1 & FULL == 0 & ONE_HOT_DRUGS == 0] +cnv_sum_lds_gnn <- all_param_combos[DATA_TYPES == "gnndrug cnv" & PRETRAIN == 0 & (MERGE_METHOD == "sum") & LOSS_TYPE == "weighted_rmse" & + SUBSET_TYPE == "cell_line" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +cnv_lmf_base_gnn <- all_param_combos[DATA_TYPES == "gnndrug cnv" & PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "rmse" & + SUBSET_TYPE == "drug" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +hist_rppa_trifecta <- all_param_combos[DATA_TYPES == "gnndrug hist rppa" & PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & + SUBSET_TYPE %in% c("cell_line", "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +mirna_hist_trifecta <- all_param_combos[DATA_TYPES == "gnndrug mirna hist" & PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & + SUBSET_TYPE %in% c("cell_line") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +mirna_lmf_base_gnn <- all_param_combos[DATA_TYPES == "gnndrug mirna" & PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "rmse" & + SUBSET_TYPE == "drug" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +mirna_rppa_trifecta <- all_param_combos[DATA_TYPES == "gnndrug mirna rppa" & PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & + SUBSET_TYPE %in% c("cell_line", "drug") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +mut_cnv_trifecta <- all_param_combos[DATA_TYPES == "gnndrug mut cnv" & PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & + SUBSET_TYPE %in% c("both") & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +mut_base_lds_gnn <- all_param_combos[DATA_TYPES == "gnndrug mut" & PRETRAIN == 0 & (MERGE_METHOD == "concat") & LOSS_TYPE == "weighted_rmse" & + SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +prot_lmf_base_gnn <- all_param_combos[DATA_TYPES == "gnndrug prot" & PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "rmse" & + SUBSET_TYPE == "drug" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] + +stragglers <- rbindlist( + list( + cnv_metab_base, + cnv_sum_lds_gnn, + cnv_lmf_base_gnn, + hist_rppa_trifecta, + mirna_hist_trifecta, + mirna_lmf_base_gnn, + mirna_rppa_trifecta, + mut_cnv_trifecta, + mut_base_lds_gnn, + prot_lmf_base_gnn + ) +) + +# combos <- all_param_combos[PRETRAIN == 0 & (MERGE_METHOD == "lmf") & LOSS_TYPE == "weighted_rmse" & SUBSET_TYPE == "both" & BOTTLENECK == 0 & FULL == 0 & ONE_HOT_DRUGS == 0] +stragglers <- stragglers[, !c("ENCODER_TRAIN", "ONE_HOT_DRUGS")] +fwrite(combos, "DRP/slurm/grids/drp_opt_stragglers.csv", col.names = F)