config.json

{
  "task_description": "Your task is to develop a predictive model that accurately estimates gene expression profiles of individual K562 cells following CRISPR interference (CRISPRi), using the dataset from Norman et al. (2019, Science).\n\nTask Definition:\n- Input: Baseline gene expression profile of an unperturbed K562 cell and the identity of the target gene(s) for perturbation\n- Output: Predicted gene expression profile after perturbation\n\nEvaluation Scenarios:\n1. Unseen Perturbations: Predict effects of gene perturbations not present during training\n2. Unseen Cell Contexts: Predict responses in cells with gene expression profiles not observed during training\n\nEvaluation Metrics:\n- Mean Squared Error (MSE): Measures the average squared difference between predicted and observed gene expression.\n- Pearson Correlation Coefficient (PCC): Quantifies linear correlation between predicted and observed profiles.\n- R² (Coefficient of Determination): Represents the proportion of variance in the observed gene expression that can be explained by the predicted values.\n- MSE for Differentially Expressed (DE) Genes (MSE_DE): Same as MSE but computed specifically for genes identified as differentially expressed.\n- PCC for Differentially Expressed (DE) Genes (PCC_DE): Same as PCC but computed specifically for genes identified as differentially expressed.\n- R² for Differentially Expressed (DE) Genes (R2_DE): Same as R² but computed specifically for genes identified as differentially expressed.",
  "dataset_path": "data/datasets/",
  "output_dir": "data/",
  "llm_config": {
    "provider": "openai",
    "model": "gpt-4o-mini",
    "api_key": "loaded_from_env"
  },
  "workflow_phases": [
    "task_analysis",
    "method_design",
    "code_generation"
  ],
  "code_generation": {
    "backend": "codex"
  },
  "qdrant_config": {
    "host": "localhost",
    "port": 6333
  }
}