config.yaml

anndata_label_name: natural_language_annotation
num_replicates: 5
llava_projector_type: mlp2x_8t_gelu # NOTE fails if > 8 because we use a fully connected layer between all 8 tokens (i.e. having too many params)
precomputing_base_url: https://medical-epigenomics.org/papers/schaefer2024/data

paths:
  read_count_table: resources/{dataset}/read_count_table.h5ad
  raw_annotations: resources/{dataset}/annotations.tsv
  structured_annotations: results/{dataset}/annotations.json
  processed_annotations: results/{dataset}/processed_annotations.json
  processed_multi_annotations: results/{dataset}/processed_multi_annotations.json
  geneset_gmt: results/genesets/{library}.gmt
  model_processed_dataset: results/{dataset}/{model}/full_output.npz
  full_dataset: results/{dataset}/full_data.h5ad
  enrichr_terms_json: resources/enrichr_terms/terms.json
  ensembl_gene_symbol_map: resources/ensembl_gene_symbol_map.csv
  jointemb_models: results/models/jointemb
  wandb_logs: results/wandb_logging
  datamodule_prepared_path: results/{dataset}/lightning-processed/{hash}.pt
  cache: results/cache  # overwritable with CELLWHISPERER_CACHE env

  gsva:
    result: results/{dataset}/gsva.parquet
    correlation: results/{dataset}/{model}/gsva_correlations.csv
    cw_transcriptome_term_scores: results/{dataset}/{model}/cw_transcriptome_term_scores.parquet
    plots: results/plots/gsva_correlation/{dataset}/{model}

  zero_shot_validation:
    result_dir: results/plots/zero_shot_validation/{model,[^/]+}/

  llava:
    root: results/llava/
    pretrain_text_dataset: results/llava/pretrain_texts.json
    finetune_text_dataset: results/llava/finetune_conversations.json
    evaluation_text_dataset: results/llava/evaluation/{dataset}_conversations.json
    combined_processed_data: results/llava/combined_processed_data/{model}.npz
    pretrained_model_dir: results/llava/pretrained/{base_model}__{model}/
    finetuned_model_dir: results/llava/finetuned/{base_model}__{model}/
    evaluation_results:  results/plots/llava/{dataset}/{base_model}__{model}/
  geo_umap: results/plots/geo_umap/{model}/
  ablations:
    models: results/models/ablations/
    plots: results/plots/ablations/

plot_style: "src/plot_style/main.style"

datasets:
  - immgen
  - human_disease
  - tabula_sapiens
  - tabula_sapiens_100_cells_per_type
  - archs4_geo
  - bowel_disease
  - pancreas

zero_shot_validation_transcriptome_model_name: "geneformer"

zero_shot_validation_datasets:
  - immgen
  - human_disease
  - tabula_sapiens
  - pancreas
  - tabula_sapiens_well_studied_celltypes

metadata_cols_per_zero_shot_validation_dataset:
  immgen:
    - celltype
  human_disease:
    - Disease_subtype
    - Tissue
  pancreas:
    - celltype
  tabula_sapiens:
    - celltype
    - organ_tissue
  tabula_sapiens_well_studied_celltypes:
    - celltype

retrieval_validation_sets:
  - immgen_deduplicated
  - human_disease_strictly_deduplicated_dmis-lab_biobert-v1.1_CLS_pooling

model_name_path_map:
  geneformer: resources/geneformer-12L-30M
  biogpt: microsoft/biogpt
  bert: dmis-lab/biobert-v1.1
  scgpt: resources/scGPT_human
  cellwhisperer: cellwhisperer_clip_v1  # wandb run ID t9kkor63
  llava_base_llm: Mistral-7B-Instruct-v0.2
  mixtral: resources/mixtral-8x7b-instruct-v0.1.Q5_K_M.gguf

genesets:
  - "Azimuth_2023"
  - "GO_Biological_Process_2023"
  - "GO_Molecular_Function_2023"
  - "OMIM_Expanded"
  - "PanglaoDB_Augmented_2021"
  - "Tabula_Sapiens"

# Created by running this on the full tabula sapiens dataset: list(adata.obs[adata.obs["organ_tissue"].isin(["Liver","Lung","Blood"])]["cell_ontology_class"].value_counts().iloc[:20].index)
top20_lung_liver_blood_celltypes:
  - "macrophage"
  - "erythrocyte"
  - "monocyte"
  - "type ii pneumocyte"
  - "classical monocyte"
  - "neutrophil"
  - "cd4-positive, alpha-beta t cell"
  - "nk cell"
  - "naive b cell"
  - "basal cell"
  - "cd8-positive, alpha-beta t cell"
  - "hepatocyte"
  - "cd8-positive, alpha-beta cytokine secreting effector t cell"
  - "club cell"
  - "non-classical monocyte"
  - "capillary endothelial cell"
  - "cd4-positive, alpha-beta memory t cell"
  - "memory b cell"
  - "respiratory goblet cell"
  - "basophil"