feat: new dataset structure

LOADING_TROUBLESHOOTING.md

@@ -1,49 +0,0 @@
-# Loading Troubleshooting
-
-## CastError: Column Names Don't Match
-
-If you encounter a `CastError` about column names not matching when loading the dataset from Hugging Face Hub, this is caused by schema conflicts between different file formats in the repository.
-
-### Quick Fix
-
-Use explicit JSONL data files:
-
-```python
-from datasets import load_dataset
-
-dataset = load_dataset('json', data_files={
-    'train': 'hf://datasets/shlokn/autogkb/train.jsonl',
-    'validation': 'hf://datasets/shlokn/autogkb/val.jsonl', 
-    'test': 'hf://datasets/shlokn/autogkb/test.jsonl'
-})
-```
-
-### Why This Happens
-
-The dataset repository contains both:
-- `.jsonl` files with the processed dataset (recommended for most use cases)
-- `.tsv` files with raw annotations (for manual inspection)
-
-The Hugging Face datasets library auto-detects both formats and tries to merge them, but they have slightly different column names:
-- JSONL files: `isplural`, `is_plural` (processed field names)
-- TSV files: `isPlural` (original column name)
-
-### The Fix
-
-We've added a `.huggingfaceignore` file to prevent auto-detection of the TSV files. However, it may take some time for this change to propagate on the Hub.
-
-### Alternative Loading Methods
-
-See `load_dataset_examples.py` for multiple ways to load the dataset:
-1. **Explicit JSONL files** (recommended, most reliable)
-2. **Standard HF loading** (works after .huggingfaceignore is processed)  
-3. **Local loading** (if you have the repo cloned)
-4. **Manual loading** (using pandas directly)
-
-### Verification
-
-Run the test script to verify loading works:
-
-```bash
-python test_hf_loading.py
-```

README.md

@@ -20,8 +20,16 @@ multilinguality:
 pretty_name: AutoGKB Annotation Benchmark
 dataset_info:
   features:
-  - name: variant_annotation_id
+  - name: pmcid
+    dtype: string
+  - name: article_title
+    dtype: string
+  - name: article_path
+    dtype: string
+  - name: article_text
     dtype: string
+  - name: variant_annotation_id
+    dtype: int64
   - name: variant_haplotypes
     dtype: string
   - name: gene
@@ -29,7 +37,7 @@ dataset_info:
   - name: drugs
     dtype: string
   - name: pmid
-    dtype: string
+    dtype: int64
   - name: phenotype_category
     dtype: string
   - name: significance
@@ -64,8 +72,6 @@ dataset_info:
     dtype: string
   - name: comparison_metabolizer_types
     dtype: string
-  - name: text
-    dtype: string
   splits:
   - name: train
     num_examples: 3124
@@ -102,25 +108,33 @@ Each example contains:
 Example:
 ```json
 {
-  "variant_annotation_id": "1450936460",
-  "variant_haplotypes": "rs28362731",
-  "gene": "AQP1",
-  "drugs": "cisplatin",
-  "pmid": "30840592",
-  "phenotype_category": "Efficacy",
-  "significance": "no",
-  "sentence": "Genotype AG is not associated with response to cisplatin in people with Mesothelioma as compared to genotype GG.",
-  "is_is_not_associated": "Not associated with",
-  "direction_of_effect": "",
-  "population_phenotypes_or_diseases": "Other:Mesothelioma",
-  "comparison_alleles_or_genotypes": "GG",
-  "text": "# Prediction of CYP2D6 poor metabolizers..."
+  "pmcid": "PMC6714673",
+  "article_title": "Warfarin Dose Model for the Prediction of Stable Maintenance Dose in Indian Patients",
+  "article_path": "articles/PMC6714673.md",
+  "article_text": "# Warfarin Dose Model for the Prediction of Stable Maintenance Dose in Indian Patients\n\n## Abstract\n\nWarfarin is a commonly used anticoagulant...",
+  "variant_annotation_id": 1449192282,
+  "variant_haplotypes": "rs1799853",
+  "gene": "CYP2C9",
+  "drugs": "warfarin",
+  "pmid": 28049362,
+  "phenotype_category": "Dosage",
+  "significance": "yes",
+  "sentence": "Genotype CT is associated with decreased dose of warfarin as compared to genotype CC.",
+  "alleles": "CT",
+  "is_is_not_associated": "Associated with",
+  "direction_of_effect": "decreased",
+  "pd_pk_terms": "dose of",
+  "comparison_alleles_or_genotypes": "CC"
 }
 ```
 
 ### Data Fields
 
 #### Core Fields
+- `pmcid`: PubMed Central identifier of the source article
+- `article_title`: Title of the source scientific article
+- `article_path`: Relative path to the article file (markdown format)
+- `article_text`: Full text of the scientific article in markdown format
 - `variant_annotation_id`: Unique identifier for each annotation
 - `variant_haplotypes`: Genetic variant identifier (e.g., rs numbers, haplotypes)
 - `gene`: Gene symbol (e.g., CYP2D6, ABCB1)
@@ -149,8 +163,6 @@ Example:
 - `comparison_alleles_or_genotypes`: Reference genotypes for comparison
 - `comparison_metabolizer_types`: Reference metabolizer types
 
-#### Text Data
-- `text`: Full text of the source scientific article in markdown format
 
 ### Data Splits
 
@@ -234,10 +246,10 @@ This dataset is released under the Apache License 2.0.
 ### Citation Information
 
 ```bibtex
-@misc{variant_drug_benchmark_2024,
-  title={Variant-Drug Annotation Benchmark},
-  author={AutoGKB Team},
-  year={2024},
+@misc{autogkb_annotation_benchmark_2025,
+  title={AutoGKB Annotation Benchmark},
+  author={Shlok Natarajan},
+  year={2025},
   note={A benchmark for pharmacogenomic variant-drug annotation extraction from scientific literature}
 }
 ```
@@ -257,7 +269,8 @@ This dataset contributes to the biomedical NLP community by providing:
 ```python
 from datasets import load_dataset
 
-dataset = load_dataset("variant_drug_benchmark")
+# Load the dataset from Hugging Face Hub
+dataset = load_dataset("autogkb/autogkb-annotation-benchmark")
 
 # Access different splits
 train_data = dataset["train"]
@@ -268,6 +281,14 @@ test_data = dataset["test"]
 efficacy_examples = train_data.filter(
     lambda x: "Efficacy" in x["phenotype_category"]
 )
+
+# Example: Access article text for a specific annotation
+first_example = train_data[0]
+print(f"PMC ID: {first_example['pmcid']}")
+print(f"Article Title: {first_example['article_title']}")
+print(f"Gene: {first_example['gene']}")
+print(f"Drug: {first_example['drugs']}")
+print(f"Full Article Text: {first_example['article_text'][:500]}...")
 ```
 
 ### Evaluation
@@ -288,21 +309,17 @@ python evaluate.py baseline_predictions.tsv val/annotations.tsv --output results
 ## File Structure
 
 ```
-benchmark/
-├── train/
-│   ├── annotations.tsv    # Training annotations
-│   └── texts/            # Training text files (PMC*.md)
-├── val/
-│   ├── annotations.tsv    # Validation annotations  
-│   └── texts/            # Validation text files
-├── test/
-│   ├── annotations.tsv    # Test annotations
-│   └── texts/            # Test text files
-├── dataset_statistics.json
-├── evaluate.py           # Evaluation script
-├── baseline_model.py     # Rule-based baseline
-├── variant_drug_benchmark.py  # HuggingFace dataset script
-├── dataset_infos.json    # Dataset metadata
+autogkb/
+├── articles/             # Full article texts in markdown format
+│   ├── PMC10038974.md
+│   ├── PMC10085626.md
+│   └── ...              # 1,431 articles total
+├── train.jsonl          # Training annotations (3,124 examples)
+├── val.jsonl            # Validation annotations (796 examples)
+├── test.jsonl           # Test annotations (596 examples)
+├── autogkb_annotation_benchmark.py  # HuggingFace dataset script
+├── dataset_infos.json   # Dataset metadata
+├── dataset_statistics.json  # Dataset statistics
 ├── LICENSE              # Apache 2.0 license
 └── README.md           # This file
 ```

TESTING.md

@@ -1,173 +0,0 @@
-# AutoGKB Dataset Testing Guide
-
-This directory contains comprehensive testing utilities for the AutoGKB Annotation Benchmark dataset to ensure reliable loading and data integrity.
-
-## Quick Start
-
-```bash
-# Run quick test to verify everything works
-python load_dataset_safe.py test
-
-# Run comprehensive simple tests  
-python test_dataset_simple.py
-
-# Run full test suite (may take longer)
-python test_dataset_loading.py
-```
-
-## Test Files
-
-### `test_dataset_simple.py`
-**Recommended for routine testing**
-- Fast execution (< 30 seconds)
-- Tests TSV structure and headers
-- Tests dataset loading with small samples
-- Tests data content quality
-- Good for CI/CD pipelines
-
-### `test_dataset_loading.py` 
-**Comprehensive test suite**
-- Full unittest framework
-- Tests cache handling
-- Tests multiple loading methods
-- Tests data integrity across all splits
-- More thorough but slower
-
-### `load_dataset_safe.py`
-**Production-ready loading utility**
-- Handles cache issues automatically
-- Multiple fallback loading methods
-- Safe for use in development and production
-- Includes quick testing function
-
-## Common Issues and Solutions
-
-### Cache Issues
-The HuggingFace datasets library can have caching conflicts. Our utilities automatically:
-- Disable caching when needed
-- Clear problematic cache directories
-- Provide fallback loading methods
-
-### Loading Methods
-The safe loader tries multiple approaches:
-1. **Manual generation**: Direct data loading without caching
-2. **Temporary disk**: Save to temp location and reload
-3. **CSV fallback**: Direct pandas loading as last resort
-
-## Usage Examples
-
-### Basic Loading
-```python
-from load_dataset_safe import load_autogkb_safe
-
-# Load small sample for testing
-dataset = load_autogkb_safe(split="train", max_examples=10)
-
-# Load specific split
-train_data = load_autogkb_safe(split="train")
-
-# Load all splits
-dataset_dict = load_autogkb_safe()
-```
-
-### Advanced Options
-```python
-# Disable caching (recommended for development)
-dataset = load_autogkb_safe(use_cache=False)
-
-# Load from specific directory
-dataset = load_autogkb_safe(data_dir="/path/to/dataset")
-
-# Quick test function
-from load_dataset_safe import quick_test
-success = quick_test(max_examples=5)
-```
-
-## Dataset Validation
-
-The tests verify:
-
-### ✅ File Structure
-- All required TSV files exist (`train/`, `val/`, `test/`)
-- Correct column headers (22 columns matching schema)
-- Non-empty files
-
-### ✅ Data Loading
-- Dataset builder initialization
-- Manual data generation
-- Cache-aware loading
-- Fallback methods
-
-### ✅ Data Integrity  
-- No duplicate IDs within splits
-- Required fields are non-empty
-- Correct data types (all strings)
-- PMID format validation
-
-### ✅ Text Matching
-- PMID to text file matching
-- Text content availability
-- Cross-reference validation
-
-## Expected Test Output
-
-```
-🧪 AutoGKB Dataset Simple Test Suite
-==================================================
-🔍 Testing TSV file structure...
-  ✅ train/annotations.tsv: 22 columns
-  ✅ val/annotations.tsv: 22 columns  
-  ✅ test/annotations.tsv: 22 columns
-✅ TSV structure test passed!
-
-📦 Testing dataset loading...
-    ✅ train: 3 examples loaded
-    ✅ validation: 3 examples loaded
-    ✅ test: 3 examples loaded
-✅ Dataset loading test passed!
-
-📚 Testing full dataset loading...
-✅ Full dataset test passed!
-  Splits: ['train', 'validation', 'test']
-  train: 5 examples
-  validation: 5 examples
-  test: 5 examples
-
-🔍 Testing data content quality...
-✅ Data content test passed! Checked 10 examples
-
-🎉 All tests passed!
-```
-
-## Troubleshooting
-
-### If tests fail:
-
-1. **Check file structure**: Ensure `train/`, `val/`, `test/` directories exist with `annotations.tsv` files
-2. **Clear cache**: Run `python -c "from load_dataset_safe import clear_datasets_cache; clear_datasets_cache()"`
-3. **Check dependencies**: Ensure `datasets`, `pandas` are installed
-4. **Run with verbose output**: Add print statements or use the comprehensive test suite
-
-### Common error messages:
-
-- **"Required file missing"**: Check that TSV files exist in correct locations
-- **"Column mismatch"**: Verify TSV headers match schema (run column header test)
-- **"Cache issues"**: Use `use_cache=False` or clear cache directory
-- **"Loading failed"**: Try fallback methods or check file permissions
-
-## Integration with CI/CD
-
-For automated testing:
-
-```bash
-# Add to your CI pipeline
-python test_dataset_simple.py
-if [ $? -eq 0 ]; then
-    echo "Dataset tests passed"
-else
-    echo "Dataset tests failed"
-    exit 1
-fi
-```
-
-The simple test suite is designed to be fast and reliable for continuous integration environments.