Mercurial > repos > bimib > cobraxy
comparison COBRAxy/docs/tutorials/python-api.md @ 492:4ed95023af20 draft
Uploaded
author | francesco_lapi |
---|---|
date | Tue, 30 Sep 2025 14:02:17 +0000 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
491:7a413a5ec566 | 492:4ed95023af20 |
---|---|
1 # Python API Tutorial | |
2 | |
3 Learn how to use COBRAxy tools programmatically in Python scripts and analysis pipelines. | |
4 | |
5 ## Overview | |
6 | |
7 This tutorial teaches you to integrate COBRAxy into Python workflows by calling tool main functions directly with parsed arguments. | |
8 | |
9 **Time required**: ~45 minutes | |
10 **Difficulty**: Intermediate | |
11 **Prerequisites**: Basic Python knowledge, COBRAxy installation | |
12 | |
13 ## Understanding COBRAxy Architecture | |
14 | |
15 ### Tool Structure | |
16 | |
17 Each COBRAxy tool is a Python module with: | |
18 - `main(args)` function that accepts argument list | |
19 - Command-line argument parsing | |
20 - Self-contained execution logic | |
21 | |
22 ```python | |
23 # General pattern for all tools | |
24 import tool_module | |
25 tool_module.main(['-arg1', 'value1', '-arg2', 'value2']) | |
26 ``` | |
27 | |
28 ### Available Tools | |
29 | |
30 | Python Module | Purpose | Key Arguments | | |
31 |---------------|---------|---------------| | |
32 | `ras_generator` | Compute reaction activity scores | `-in`, `-ra`, `-rs` | | |
33 | `rps_generator` | Compute reaction propensity scores | `-id`, `-rp` | | |
34 | `marea` | Statistical pathway analysis | `-input_data`, `-choice_map` | | |
35 | `ras_to_bounds` | Apply RAS constraints to model | `-ir`, `-ms`, `-idop` | | |
36 | `flux_simulation` | Sample metabolic fluxes | `-ms`, `-in`, `-a`, `-ns` | | |
37 | `flux_to_map` | Add flux data to maps | `-if`, `-mp`, `-idop` | | |
38 | |
39 ## Setup Your Environment | |
40 | |
41 ### Import Required Modules | |
42 | |
43 ```python | |
44 import sys | |
45 import os | |
46 from pathlib import Path | |
47 | |
48 # Add COBRAxy to Python path | |
49 cobraxy_path = "/path/to/COBRAxy" | |
50 sys.path.insert(0, cobraxy_path) | |
51 | |
52 # Import COBRAxy tools | |
53 import ras_generator | |
54 import rps_generator | |
55 import marea | |
56 import ras_to_bounds | |
57 import flux_simulation | |
58 import flux_to_map | |
59 import metabolicModel2Tabular as model_setting | |
60 ``` | |
61 | |
62 ### Set Working Directory | |
63 | |
64 ```python | |
65 # Set up working directory | |
66 work_dir = Path("/path/to/analysis") | |
67 work_dir.mkdir(exist_ok=True) | |
68 os.chdir(work_dir) | |
69 | |
70 # COBRAxy tools expect this parameter | |
71 tool_dir = str(Path(cobraxy_path).absolute()) | |
72 ``` | |
73 | |
74 ## Basic Workflow Example | |
75 | |
76 ### Step 1: Prepare Sample Data | |
77 | |
78 ```python | |
79 import pandas as pd | |
80 import numpy as np | |
81 | |
82 # Create sample gene expression data | |
83 genes = ['HGNC:5', 'HGNC:10', 'HGNC:15', 'HGNC:25', 'HGNC:30'] | |
84 samples = ['Control_1', 'Control_2', 'Treatment_1', 'Treatment_2'] | |
85 | |
86 # Generate random expression values | |
87 np.random.seed(42) | |
88 data = np.random.lognormal(mean=2, sigma=1, size=(len(genes), len(samples))) | |
89 | |
90 # Create DataFrame | |
91 expression_df = pd.DataFrame(data, index=genes, columns=samples) | |
92 expression_df.index.name = 'Gene_ID' | |
93 | |
94 # Save to file | |
95 expression_file = work_dir / "expression_data.tsv" | |
96 expression_df.to_csv(expression_file, sep='\t') | |
97 print(f"Created sample data: {expression_file}") | |
98 ``` | |
99 | |
100 ### Step 2: Extract Model Information | |
101 | |
102 ```python | |
103 # Extract model components (optional, for understanding model structure) | |
104 model_args = [ | |
105 '-td', tool_dir, | |
106 '-ms', 'ENGRO2', # Use built-in ENGRO2 model | |
107 '-idop', str(work_dir / 'model_info') | |
108 ] | |
109 | |
110 try: | |
111 model_setting.main(model_args) | |
112 print("✓ Model information extracted") | |
113 except Exception as e: | |
114 print(f"Model extraction failed: {e}") | |
115 ``` | |
116 | |
117 ### Step 3: Generate RAS Scores | |
118 | |
119 ```python | |
120 # Generate Reaction Activity Scores | |
121 ras_output = work_dir / "ras_scores.tsv" | |
122 | |
123 ras_args = [ | |
124 '-td', tool_dir, | |
125 '-in', str(expression_file), | |
126 '-ra', str(ras_output), | |
127 '-rs', 'ENGRO2', # Built-in model | |
128 '-n', 'true' # Handle missing genes | |
129 ] | |
130 | |
131 try: | |
132 ras_generator.main(ras_args) | |
133 print(f"✓ RAS scores generated: {ras_output}") | |
134 except Exception as e: | |
135 print(f"RAS generation failed: {e}") | |
136 raise | |
137 ``` | |
138 | |
139 ### Step 4: Generate RPS Scores (Optional) | |
140 | |
141 ```python | |
142 # Create sample metabolite data | |
143 metabolites = ['glucose', 'pyruvate', 'lactate', 'ATP', 'NADH'] | |
144 met_data = np.random.lognormal(mean=3, sigma=0.5, size=(len(metabolites), len(samples))) | |
145 | |
146 met_df = pd.DataFrame(met_data, index=metabolites, columns=samples) | |
147 met_df.index.name = 'Metabolite_ID' | |
148 | |
149 metabolite_file = work_dir / "metabolite_data.tsv" | |
150 met_df.to_csv(metabolite_file, sep='\t') | |
151 | |
152 # Generate Reaction Propensity Scores | |
153 rps_output = work_dir / "rps_scores.tsv" | |
154 | |
155 rps_args = [ | |
156 '-td', tool_dir, | |
157 '-id', str(metabolite_file), | |
158 '-rp', str(rps_output) | |
159 ] | |
160 | |
161 try: | |
162 rps_generator.main(rps_args) | |
163 print(f"✓ RPS scores generated: {rps_output}") | |
164 except Exception as e: | |
165 print(f"RPS generation warning: {e}") | |
166 # RPS generation might fail with sample data - that's OK | |
167 ``` | |
168 | |
169 ### Step 5: Statistical Analysis with MAREA | |
170 | |
171 ```python | |
172 # Create enriched pathway maps | |
173 maps_output = work_dir / "pathway_maps" | |
174 | |
175 marea_args = [ | |
176 '-td', tool_dir, | |
177 '-using_RAS', 'true', | |
178 '-input_data', str(ras_output), | |
179 '-choice_map', 'ENGRO2', | |
180 '-gs', 'true', # Gene set analysis | |
181 '-idop', str(maps_output) | |
182 ] | |
183 | |
184 try: | |
185 marea.main(marea_args) | |
186 print(f"✓ Pathway maps created: {maps_output}") | |
187 except Exception as e: | |
188 print(f"MAREA analysis failed: {e}") | |
189 ``` | |
190 | |
191 ### Step 6: Flux Simulation Pipeline | |
192 | |
193 ```python | |
194 # Apply RAS constraints to model | |
195 bounds_output = work_dir / "model_bounds" | |
196 | |
197 bounds_args = [ | |
198 '-td', tool_dir, | |
199 '-ms', 'ENGRO2', | |
200 '-ir', str(ras_output), | |
201 '-rs', 'true', # Use RAS values | |
202 '-idop', str(bounds_output) | |
203 ] | |
204 | |
205 try: | |
206 ras_to_bounds.main(bounds_args) | |
207 print(f"✓ Model constraints applied: {bounds_output}") | |
208 except Exception as e: | |
209 print(f"Bounds generation failed: {e}") | |
210 raise | |
211 | |
212 # Sample metabolic fluxes | |
213 flux_output = work_dir / "flux_samples" | |
214 | |
215 flux_args = [ | |
216 '-td', tool_dir, | |
217 '-ms', 'ENGRO2', | |
218 '-in', str(bounds_output / "*.tsv"), # Will be expanded by tool | |
219 '-a', 'CBS', # Sampling algorithm | |
220 '-ns', '1000', # Number of samples | |
221 '-idop', str(flux_output) | |
222 ] | |
223 | |
224 try: | |
225 flux_simulation.main(flux_args) | |
226 print(f"✓ Flux samples generated: {flux_output}") | |
227 except Exception as e: | |
228 print(f"Flux simulation failed: {e}") | |
229 ``` | |
230 | |
231 ### Step 7: Create Final Visualizations | |
232 | |
233 ```python | |
234 # Add flux data to enriched maps | |
235 final_maps = work_dir / "final_visualizations" | |
236 | |
237 # Check if we have both maps and flux data | |
238 maps_dir = maps_output | |
239 flux_dir = flux_output | |
240 | |
241 if maps_dir.exists() and flux_dir.exists(): | |
242 flux_to_map_args = [ | |
243 '-td', tool_dir, | |
244 '-if', str(flux_dir / "*.tsv"), | |
245 '-mp', str(maps_dir / "*.svg"), | |
246 '-idop', str(final_maps) | |
247 ] | |
248 | |
249 try: | |
250 flux_to_map.main(flux_to_map_args) | |
251 print(f"✓ Final visualizations created: {final_maps}") | |
252 except Exception as e: | |
253 print(f"Final mapping failed: {e}") | |
254 else: | |
255 print("Skipping final visualization - missing input files") | |
256 ``` | |
257 | |
258 ## Advanced Usage Patterns | |
259 | |
260 ### Error Handling and Validation | |
261 | |
262 ```python | |
263 def run_cobraxy_tool(tool_module, args, description): | |
264 """Helper function to run COBRAxy tools with error handling.""" | |
265 try: | |
266 print(f"Running {description}...") | |
267 tool_module.main(args) | |
268 print(f"✓ {description} completed successfully") | |
269 return True | |
270 except Exception as e: | |
271 print(f"✗ {description} failed: {e}") | |
272 return False | |
273 | |
274 # Usage | |
275 success = run_cobraxy_tool( | |
276 ras_generator, | |
277 ras_args, | |
278 "RAS generation" | |
279 ) | |
280 | |
281 if not success: | |
282 print("Pipeline stopped due to error") | |
283 exit(1) | |
284 ``` | |
285 | |
286 ### Batch Processing Multiple Datasets | |
287 | |
288 ```python | |
289 def process_dataset(dataset_path, output_dir): | |
290 """Process a single dataset through COBRAxy pipeline.""" | |
291 | |
292 dataset_name = dataset_path.stem | |
293 out_dir = Path(output_dir) / dataset_name | |
294 out_dir.mkdir(exist_ok=True) | |
295 | |
296 # Generate RAS | |
297 ras_file = out_dir / "ras_scores.tsv" | |
298 ras_args = [ | |
299 '-td', tool_dir, | |
300 '-in', str(dataset_path), | |
301 '-ra', str(ras_file), | |
302 '-rs', 'ENGRO2' | |
303 ] | |
304 | |
305 if run_cobraxy_tool(ras_generator, ras_args, f"RAS for {dataset_name}"): | |
306 # Continue with MAREA analysis | |
307 maps_dir = out_dir / "maps" | |
308 marea_args = [ | |
309 '-td', tool_dir, | |
310 '-using_RAS', 'true', | |
311 '-input_data', str(ras_file), | |
312 '-choice_map', 'ENGRO2', | |
313 '-idop', str(maps_dir) | |
314 ] | |
315 run_cobraxy_tool(marea, marea_args, f"MAREA for {dataset_name}") | |
316 | |
317 return out_dir | |
318 | |
319 # Process multiple datasets | |
320 datasets = [ | |
321 "/path/to/dataset1.tsv", | |
322 "/path/to/dataset2.tsv", | |
323 "/path/to/dataset3.tsv" | |
324 ] | |
325 | |
326 results = [] | |
327 for dataset in datasets: | |
328 result_dir = process_dataset(Path(dataset), work_dir / "batch_results") | |
329 results.append(result_dir) | |
330 | |
331 print(f"Processed {len(results)} datasets") | |
332 ``` | |
333 | |
334 ### Custom Analysis Pipelines | |
335 | |
336 ```python | |
337 class COBRAxyPipeline: | |
338 """Custom COBRAxy analysis pipeline.""" | |
339 | |
340 def __init__(self, tool_dir, work_dir): | |
341 self.tool_dir = tool_dir | |
342 self.work_dir = Path(work_dir) | |
343 self.work_dir.mkdir(exist_ok=True) | |
344 | |
345 def run_enrichment_analysis(self, expression_file, model='ENGRO2'): | |
346 """Run enrichment-focused analysis.""" | |
347 | |
348 # Generate RAS | |
349 ras_file = self.work_dir / "ras_scores.tsv" | |
350 ras_args = ['-td', self.tool_dir, '-in', str(expression_file), | |
351 '-ra', str(ras_file), '-rs', model] | |
352 | |
353 if not run_cobraxy_tool(ras_generator, ras_args, "RAS generation"): | |
354 return None | |
355 | |
356 # Run MAREA | |
357 maps_dir = self.work_dir / "enrichment_maps" | |
358 marea_args = ['-td', self.tool_dir, '-using_RAS', 'true', | |
359 '-input_data', str(ras_file), '-choice_map', model, | |
360 '-gs', 'true', '-idop', str(maps_dir)] | |
361 | |
362 if run_cobraxy_tool(marea, marea_args, "MAREA analysis"): | |
363 return maps_dir | |
364 return None | |
365 | |
366 def run_flux_analysis(self, expression_file, model='ENGRO2', n_samples=1000): | |
367 """Run flux sampling analysis.""" | |
368 | |
369 # Generate RAS and apply bounds | |
370 ras_file = self.work_dir / "ras_scores.tsv" | |
371 bounds_dir = self.work_dir / "bounds" | |
372 flux_dir = self.work_dir / "flux_samples" | |
373 | |
374 # RAS generation | |
375 ras_args = ['-td', self.tool_dir, '-in', str(expression_file), | |
376 '-ra', str(ras_file), '-rs', model] | |
377 if not run_cobraxy_tool(ras_generator, ras_args, "RAS generation"): | |
378 return None | |
379 | |
380 # Apply bounds | |
381 bounds_args = ['-td', self.tool_dir, '-ms', model, '-ir', str(ras_file), | |
382 '-rs', 'true', '-idop', str(bounds_dir)] | |
383 if not run_cobraxy_tool(ras_to_bounds, bounds_args, "Bounds application"): | |
384 return None | |
385 | |
386 # Flux sampling | |
387 flux_args = ['-td', self.tool_dir, '-ms', model, | |
388 '-in', str(bounds_dir / "*.tsv"), | |
389 '-a', 'CBS', '-ns', str(n_samples), '-idop', str(flux_dir)] | |
390 | |
391 if run_cobraxy_tool(flux_simulation, flux_args, "Flux simulation"): | |
392 return flux_dir | |
393 return None | |
394 | |
395 # Usage | |
396 pipeline = COBRAxyPipeline(tool_dir, work_dir / "custom_analysis") | |
397 | |
398 # Run enrichment analysis | |
399 enrichment_results = pipeline.run_enrichment_analysis(expression_file) | |
400 if enrichment_results: | |
401 print(f"Enrichment analysis completed: {enrichment_results}") | |
402 | |
403 # Run flux analysis | |
404 flux_results = pipeline.run_flux_analysis(expression_file, n_samples=500) | |
405 if flux_results: | |
406 print(f"Flux analysis completed: {flux_results}") | |
407 ``` | |
408 | |
409 ## Integration with Data Analysis Libraries | |
410 | |
411 ### Pandas Integration | |
412 | |
413 ```python | |
414 # Read COBRAxy results back into pandas | |
415 ras_df = pd.read_csv(ras_output, sep='\t', index_col=0) | |
416 print(f"RAS data shape: {ras_df.shape}") | |
417 print(f"Sample statistics:\n{ras_df.describe()}") | |
418 | |
419 # Filter highly variable reactions | |
420 ras_std = ras_df.std(axis=1) | |
421 variable_reactions = ras_std.nlargest(20).index | |
422 print(f"Most variable reactions: {list(variable_reactions)}") | |
423 ``` | |
424 | |
425 ### Matplotlib Visualization | |
426 | |
427 ```python | |
428 import matplotlib.pyplot as plt | |
429 import seaborn as sns | |
430 | |
431 # Visualize RAS distributions | |
432 plt.figure(figsize=(12, 8)) | |
433 sns.heatmap(ras_df.iloc[:50], cmap='RdBu_r', center=0, cbar_kws={'label': 'RAS Score'}) | |
434 plt.title('Reaction Activity Scores (Top 50 Reactions)') | |
435 plt.xlabel('Samples') | |
436 plt.ylabel('Reactions') | |
437 plt.tight_layout() | |
438 plt.savefig(work_dir / 'ras_heatmap.png', dpi=300) | |
439 plt.show() | |
440 ``` | |
441 | |
442 ## Best Practices | |
443 | |
444 ### 1. Environment Management | |
445 ```python | |
446 # Use pathlib for cross-platform compatibility | |
447 from pathlib import Path | |
448 | |
449 # Use absolute paths | |
450 tool_dir = str(Path(cobraxy_path).absolute()) | |
451 work_dir = Path("/analysis").absolute() | |
452 ``` | |
453 | |
454 ### 2. Error Handling | |
455 ```python | |
456 # Always wrap tool calls in try-except | |
457 try: | |
458 ras_generator.main(ras_args) | |
459 except Exception as e: | |
460 print(f"RAS generation failed: {e}") | |
461 # Log details, cleanup, or alternative action | |
462 ``` | |
463 | |
464 ### 3. Argument Validation | |
465 ```python | |
466 def validate_file_exists(filepath): | |
467 """Validate input file exists.""" | |
468 path = Path(filepath) | |
469 if not path.exists(): | |
470 raise FileNotFoundError(f"Input file not found: {filepath}") | |
471 return str(path.absolute()) | |
472 | |
473 # Use before calling tools | |
474 expression_file = validate_file_exists(expression_file) | |
475 ``` | |
476 | |
477 | |
478 | |
479 ## Troubleshooting | |
480 | |
481 ### Common Issues | |
482 | |
483 **Import errors** | |
484 ```python | |
485 # Check if COBRAxy path is correct | |
486 import sys | |
487 print("Python path includes:") | |
488 for p in sys.path: | |
489 print(f" {p}") | |
490 | |
491 # Add COBRAxy path | |
492 sys.path.insert(0, "/correct/path/to/COBRAxy") | |
493 ``` | |
494 | |
495 **Tool execution failures** | |
496 ```python | |
497 # Enable verbose output | |
498 import logging | |
499 logging.basicConfig(level=logging.DEBUG) | |
500 | |
501 # Check working directory | |
502 print(f"Current directory: {os.getcwd()}") | |
503 print(f"Directory contents: {list(Path('.').iterdir())}") | |
504 ``` | |
505 | |
506 **File path issues** | |
507 ```python | |
508 # Use absolute paths | |
509 ras_args = [ | |
510 '-td', str(Path(tool_dir).absolute()), | |
511 '-in', str(Path(expression_file).absolute()), | |
512 '-ra', str(Path(ras_output).absolute()), | |
513 '-rs', 'ENGRO2' | |
514 ] | |
515 ``` | |
516 | |
517 ## Next Steps | |
518 | |
519 Now that you can use COBRAxy programmatically: | |
520 | |
521 1. **[Tools Reference](../tools/)** - Detailed parameter documentation | |
522 2. **[Examples](../examples/)** - Real-world analysis scripts | |
523 3. **Build custom analysis pipelines** for your research needs | |
524 4. **Integrate with workflow managers** like Snakemake or Nextflow | |
525 | |
526 ## Resources | |
527 | |
528 - [COBRApy Documentation](https://cobrapy.readthedocs.io/) - Underlying metabolic modeling library | |
529 - [Pandas Documentation](https://pandas.pydata.org/) - Data manipulation | |
530 - [Matplotlib Gallery](https://matplotlib.org/gallery/) - Visualization examples | |
531 - [Python Pathlib](https://docs.python.org/3/library/pathlib.html) - Modern path handling |