# =============================================================================
# SampleDisco — DEMO CONFIG (Zenodo tutorial data)
# =============================================================================
# Ready-to-run config for the demo datasets published at
#   https://doi.org/10.5281/zenodo.20988712   (concept DOI — always resolves to the latest version)
# Quick start (run from a folder that contains ./data/test_RNA.h5ad and
# ./data/test_ATAC.h5ad — see the "Demo data" tutorial page):
#   sampledisco -m complex --config <this-file>.yaml
# Metadata (sample, batch, sev.level) is embedded in each .h5ad .obs, so no
# sample-metadata CSV is required. All keys below are required and validated
# exactly against wrapper(); edit values, do not remove keys.
# =============================================================================


# =============================================================================
# MAIN CONFIGURATION
# =============================================================================
output_dir: "./sampledisco_demo_output"

run_rna_pipeline: true
run_atac_pipeline: true
run_multiomics_pipeline: false   # set true to run multi-omics; pre-wired to skip scGLUE training (see multiomics_integration below)

# General settings
use_gpu: false   # set true to use the RAPIDS GPU paths if installed; auto-falls back to CPU otherwise
initialization: false
verbose: true
save_intermediate: true
large_data_need_extra_memory: false

# =============================================================================
# RNA PIPELINE PARAMETERS
# =============================================================================

# --- Input/Output Paths ---
rna_count_data_path: "./data/test_RNA.h5ad"
rna_output_dir:       # defaults to {output_dir}/rna

# --- Pipeline Control Flags --- TEST: all on
rna_preprocessing: true
rna_cell_type_cluster: true
rna_derive_sample_embedding: true
rna_sample_distance_calculation: true
rna_trajectory_analysis: true
rna_trajectory_dge: true
rna_sample_cluster: true
rna_proportion_test: true
rna_cluster_dge: false   # RAISIN cluster-DGE is memory-heavy (multithreaded ComBat scales with #cores); off by default so the demo finishes on modest machines. Set true if you have ample RAM.
rna_visualize_data: true
rna_dimension_association_analysis: true

# --- Resume Paths (for skipping preprocessing) ---
rna_adata_cell_path:
rna_adata_sample_path:
rna_sample_meta_path:
rna_cell_meta_path:
rna_pseudo_adata_path:

# --- Common Column Names ---
rna_sample_col: "sample"
rna_sample_level_batch_col: "batch"
rna_celltype_col: "cell_type"   # where de-novo cluster labels are written (existing_cell_types:false). The demo .h5ad's precomputed labels are in 'celltype' — set this to "celltype" only if you flip existing_cell_types:true.

# --- Preprocessing Parameters ---
rna_min_cells: 100
rna_min_genes: 200
rna_pct_mito_cutoff: 20.0
rna_exclude_genes:
rna_num_cell_hvgs: 2000
rna_cell_embedding_num_pcs: 20
rna_num_harmony_iterations: 20
rna_cell_level_batch_key:
- "batch"

# --- Cell Type Clustering Parameters ---
rna_leiden_cluster_resolution: 0.8
rna_cell_embedding_column:
rna_existing_cell_types: false
rna_n_target_cell_clusters: 8
rna_umap: true

# --- Sample Embedding Parameters (composition + RMD) ---
rna_sample_embedding_medium_K: 120
rna_sample_embedding_fine_K: 300
rna_sample_embedding_rmd_dim: 8
rna_sample_embedding_use_clr: false
rna_sample_embedding_use_rmd: true
rna_sample_embedding_block_weights:
rna_sample_embedding_rmd_weight: 0.60
rna_sample_embedding_pca_components: 10
rna_sample_embedding_batch_method: "harmony"

# --- Autotune Parameters ---
rna_autotune_enable: false
rna_autotune_search: "bayesian"
rna_autotune_scoring: "auto"
rna_autotune_scope: "alpha_only"
rna_autotune_alpha_bounds: [0.1, 10.0]
rna_autotune_grouping_col:

# --- Trajectory Analysis Parameters ---
rna_n_cca_pcs: 5
rna_trajectory_col: "sev.level"
rna_trajectory_supervised: true
rna_trajectory_visualization_label:
- "sev.level"
rna_cca_pvalue: false
rna_tscan_origin:
rna_tscan_n_clusters: 3  # int; null lets TSCAN auto-select via BIC (2-20)
rna_tscan_pseudotime_mode: "rank"  # "rank" or "projection"

# --- Sample Distance Parameters ---
rna_sample_distance_methods:
- "cosine"
- "correlation"
rna_grouping_columns:
- "sev.level"
rna_summary_sample_csv_path:

# --- Trajectory Differential Gene Analysis Parameters ---
rna_fdr_threshold: 0.05
rna_effect_size_threshold: 1.0
rna_top_n_genes: 100
rna_trajectory_diff_gene_covariate:
rna_num_splines: 5
rna_spline_order: 2
rna_visualization_gene_list:

# --- Sample Clustering Parameters ---
rna_cluster_number: 3
rna_cluster_differential_gene_group_col: "sev.level"

# --- Visualization Parameters ---
rna_age_bin_size:
rna_age_column: "age"
rna_plot_dendrogram_flag: true
rna_plot_cell_type_proportions_pca_flag: true
rna_plot_cell_type_expression_umap_flag: true

# --- Dimension Association Analysis Parameters ---
# Toggle via rna_dimension_association_analysis in the Pipeline Control Flags section.
# If the *_cols lists are null, all sample-level obs columns are auto-classified
# into continuous / categorical and tested.
rna_association_continuous_cols:
rna_association_categorical_cols:
rna_association_n_permutations: 99

# =============================================================================
# ATAC PIPELINE PARAMETERS
# =============================================================================

# --- Input/Output Paths ---
atac_count_data_path: "./data/test_ATAC.h5ad"
atac_output_dir:       # defaults to {output_dir}/atac

# --- Pipeline Control Flags --- TEST: all on
atac_preprocessing: true
atac_cell_type_cluster: true
atac_derive_sample_embedding: true
atac_sample_distance_calculation: true
atac_trajectory_analysis: true
atac_trajectory_dge: false   # 8 samples × 4 splines: GAM cannot fit (too_few_samples for every gene)
atac_sample_cluster: true
atac_proportion_test: true
atac_cluster_dge: false      # disabled: 29K cells × 230K peaks × 48 cores RAISIN OOM-kills on CPU
atac_visualize_data: true
atac_dimension_association_analysis: true

# --- Resume Paths (for skipping preprocessing) ---
atac_adata_cell_path:
atac_adata_sample_path:
atac_sample_meta_path:
atac_cell_meta_path:
atac_pseudo_adata_path:

# --- Common Column Names ---
atac_sample_col: "sample"
atac_sample_level_batch_col:
atac_celltype_col: "cell_type"
atac_cell_embedding_column:

# --- ATAC-Specific Preprocessing Parameters (tuned for small test data) ---
atac_min_cells: 10
atac_min_features: 200
atac_max_features: 50000
atac_min_cells_per_sample: 50
atac_exclude_features:
atac_cell_level_batch_key:
atac_doublet_detection: false
atac_num_cell_hvfs: 10000
atac_cell_embedding_num_pcs: 30
atac_num_harmony_iterations: 20
atac_tfidf_scale_factor: 10000.0
atac_log_transform: true
atac_drop_first_lsi: true

# --- Cell Type Clustering Parameters ---
atac_leiden_cluster_resolution: 0.8
atac_existing_cell_types: false
atac_n_target_cell_clusters: 8
atac_umap: true

# --- Sample Embedding Parameters (composition + RMD) ---
atac_sample_embedding_medium_K: 120
atac_sample_embedding_fine_K: 300
atac_sample_embedding_rmd_dim: 8
atac_sample_embedding_use_clr: false
atac_sample_embedding_use_rmd: true
atac_sample_embedding_block_weights:
atac_sample_embedding_rmd_weight: 0.60
atac_sample_embedding_pca_components: 10
atac_sample_embedding_batch_method: "harmony"

# --- Autotune Parameters ---
atac_autotune_enable: false
atac_autotune_search: "bayesian"
atac_autotune_scoring: "auto"
atac_autotune_scope: "alpha_only"
atac_autotune_alpha_bounds: [0.1, 10.0]
atac_autotune_grouping_col:

# --- Trajectory Analysis Parameters ---
atac_n_cca_pcs: 5
atac_trajectory_col: "sev.level"
atac_trajectory_supervised: true
atac_trajectory_visualization_label:
- "sev.level"
atac_cca_pvalue: false
atac_tscan_origin:
atac_tscan_n_clusters: 3  # int; null lets TSCAN auto-select via BIC (2-20)
atac_tscan_pseudotime_mode: "rank"  # "rank" or "projection"

# --- Sample Distance Parameters ---
atac_sample_distance_methods:
- "cosine"
- "correlation"
atac_grouping_columns:
- "sev.level"
atac_summary_sample_csv_path:

# --- Trajectory Differential Gene Analysis Parameters ---
atac_fdr_threshold: 0.05
atac_effect_size_threshold: 1.0
atac_top_n_genes: 100
atac_trajectory_diff_gene_covariate:
atac_num_splines: 5
atac_spline_order: 3
atac_visualization_gene_list:

# --- Sample Clustering Parameters ---
atac_cluster_number: 3
atac_cluster_differential_gene_group_col: "sev.level"

# --- Visualization Parameters ---
atac_age_bin_size:
atac_age_column: "age"
atac_plot_dendrogram_flag: true
atac_plot_cell_type_proportions_pca_flag: false
atac_plot_cell_type_expression_umap_flag: false

# --- Dimension Association Analysis Parameters ---
atac_association_continuous_cols:
atac_association_categorical_cols:
atac_association_n_permutations: 99

# =============================================================================
# MULTIOMICS PIPELINE PARAMETERS
# =============================================================================

# --- Input/Output Paths ---
multiomics_rna_file: "./data/test_RNA.h5ad"
multiomics_atac_file: "./data/test_ATAC.h5ad"
multiomics_output_dir:       # defaults to {output_dir}/multiomics

# --- Pipeline Control Flags (Preprocessing) ---
multiomics_integration: false   # false -> SKIP scGLUE training, load the pre-integrated file below. Set true to train from scratch (needs bedtools) + set the resume path to null.

# --- Pipeline Control Flags (Downstream) --- TEST: all on
multiomics_sample_distance_calculation: true
multiomics_trajectory_analysis: true
multiomics_trajectory_dge: false   # 16 (sample, modality) units: GAM cannot fit on small test
multiomics_sample_cluster: true
multiomics_proportion_test: true
multiomics_cluster_dge: false      # disabled: same RAISIN OOM risk as ATAC (large peak matrix)
multiomics_visualize_embedding: true
multiomics_dimension_association_analysis: true

# --- GLUE Sub-Pipeline Flags ---
multiomics_run_glue_preprocessing: true
multiomics_run_glue_training: true
multiomics_run_glue_merge: true                    # build embedding-only union (adata_sample.h5ad)
multiomics_run_glue_preprocess_per_modality: true  # per-modality QC h5ads for DGE/RAISIN
multiomics_cell_type_cluster: true
multiomics_run_glue_visualization: true

# --- Resume Paths (for skipping steps) ---
multiomics_integrated_h5ad_path: "./data/test_multiomics_integrated.h5ad"   # scGLUE-integrated object from Zenodo; used when multiomics_integration: false (skips training)
multiomics_pseudobulk_h5ad_path:
multiomics_rna_sample_meta_file:
multiomics_atac_sample_meta_file:
multiomics_additional_hvg_file:

# --- Common Column Names ---
multiomics_rna_sample_column: "sample"
multiomics_atac_sample_column: "sample"
multiomics_sample_col: "sample"
multiomics_batch_col:
multiomics_celltype_col: "cell_type"
multiomics_modality_col: "modality"

# --- General Multiomics Settings ---
multiomics_verbose: true
multiomics_use_gpu: true
multiomics_random_state: 42

# --- GLUE Preprocessing Parameters ---
multiomics_ensembl_release: 98
multiomics_species: "homo_sapiens"
multiomics_use_highly_variable: true
multiomics_n_top_genes: 2000
multiomics_n_pca_comps: 50
multiomics_n_lsi_comps: 50
multiomics_lsi_n_iter: 15
multiomics_gtf_by: "gene_name"
multiomics_flavor: "seurat_v3"
multiomics_generate_umap: false
multiomics_compression: "gzip"

# --- GLUE Training Parameters ---
multiomics_consistency_threshold: 0.05
multiomics_treat_sample_as_batch: false  # GLUE preserves per-sample variance (V2 default)

# --- Sample-removed cluster embedding (X_glue from scGLUE is the RMD/sample-preserved emb) ---
# By default, one Harmony pass on X_glue removes per-sample variance to
# produce X_glue_harmony (used by cell typing + A1/A2/A3). Alternative:
# train scGLUE a second time with treat_sample_as_batch=True; that run's
# embedding is merged in as X_glue_harmony and the Harmony pass auto-skips.
multiomics_harmonize_xglue_max_iter: 30
multiomics_run_glue_twice_for_sample_removal: false
multiomics_save_prefix: "glue"

# scGLUE throughput knobs (test run: use scGLUE defaults; single-process loader to avoid
# empty-batch race that triggers "0x1992 vs 50x256" matmul shape error on CPU)
multiomics_glue_data_batch_size: 128
multiomics_glue_max_epochs: 50
multiomics_glue_dataloader_num_workers: 0
multiomics_glue_dataloader_fetches_per_worker: 1
multiomics_glue_array_shuffle_num_workers: 0
multiomics_glue_graph_shuffle_num_workers: 0

# --- Neighbor/Metric Parameters ---
multiomics_metric: "cosine"

# --- Cell Type Clustering Parameters ---
multiomics_existing_cell_types: false
multiomics_n_target_clusters: 8
multiomics_cluster_resolution: 0.8
# null → auto-resolve to X_glue_harmony (sample-removed) if present, else X_glue
multiomics_use_rep_celltype:
multiomics_markers:
multiomics_generate_umap_celltype: true

# --- GLUE Visualization Parameters ---
multiomics_plot_columns:

# --- Integration Preprocessing Parameters ---

# trajectory_dge / cluster_dge both off for multiomics in this test → can drop X

# --- Sample Embedding Parameters (composition + RMD on X_glue) ---
multiomics_derive_sample_embedding: true
multiomics_sample_embedding_medium_K: 120
multiomics_sample_embedding_fine_K: 300
multiomics_sample_embedding_rmd_dim: 8
multiomics_sample_embedding_use_clr: false
multiomics_sample_embedding_use_rmd: true
multiomics_sample_embedding_block_weights:
multiomics_sample_embedding_rmd_weight: 0.60
multiomics_sample_embedding_pca_components: 10
multiomics_sample_embedding_batch_method: "harmony"

# --- Autotune Parameters ---
multiomics_autotune_enable: false
multiomics_autotune_search: "bayesian"
multiomics_autotune_scoring: "auto"
multiomics_autotune_scope: "alpha_only"
multiomics_autotune_alpha_bounds: [0.1, 10.0]
multiomics_autotune_grouping_col:

# --- Trajectory Analysis Parameters ---
multiomics_trajectory_col: "sev.level"
multiomics_trajectory_supervised: true
multiomics_trajectory_visualization_label:
- "sev.level"
multiomics_n_cca_pcs: 5
multiomics_cca_pvalue: false
multiomics_tscan_origin:
multiomics_tscan_n_clusters:       # int; null lets TSCAN auto-select via BIC (2-20)
multiomics_tscan_pseudotime_mode: "rank"  # "rank" or "projection"

# --- Sample Distance Parameters ---
multiomics_sample_distance_methods:
- "cosine"
- "correlation"
multiomics_grouping_columns:
- "sev.level"
multiomics_summary_sample_csv_path:

# --- Trajectory Differential Gene Analysis Parameters ---
multiomics_fdr_threshold: 0.05
multiomics_effect_size_threshold: 1.0
multiomics_top_n_genes: 100
multiomics_trajectory_diff_gene_covariate:
multiomics_num_splines: 5
multiomics_spline_order: 3
multiomics_visualization_gene_list:

# --- Sample Clustering Parameters ---
multiomics_cluster_number: 3
multiomics_cluster_differential_gene_group_col: "sev.level"

# --- Visualization Parameters (General) ---
multiomics_age_bin_size:
multiomics_age_column: "age"
multiomics_plot_dendrogram_flag: true
multiomics_plot_cell_type_proportions_pca_flag: false
multiomics_plot_cell_type_expression_umap_flag: false

# --- Embedding Visualization Parameters ---
multiomics_color_col:
multiomics_visualization_grouping_column:
multiomics_target_modality: "ATAC"
multiomics_sample_embedding_key: "X_DR_sample"
multiomics_figsize:
- 20
- 8
multiomics_point_size: 60
multiomics_alpha: 0.8
multiomics_colormap: "viridis"
multiomics_show_sample_names: false
multiomics_force_data_type:

# --- Dimension Association Analysis Parameters ---
multiomics_association_continuous_cols:
multiomics_association_categorical_cols:
multiomics_association_n_permutations: 99
continue_on_error: false
multiomics_atac_doublet_detection: true
multiomics_atac_exclude_features:
multiomics_atac_log_transform: true
multiomics_atac_max_features: 15000
multiomics_atac_min_cells: 1
multiomics_atac_min_cells_per_sample: 1
multiomics_atac_min_features: 2000
multiomics_atac_tfidf_scale_factor: 10000.0
multiomics_autotune_tune_on_modality:
multiomics_rna_exclude_genes:
multiomics_rna_min_cells: 500
multiomics_rna_min_genes: 500
multiomics_rna_pct_mito_cutoff: 20.0
random_state: 42
multiomics_atac_min_cells_floor: 10
multiomics_n_top_peaks: 50000