File size: 31,365 Bytes
ea558c0 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 | """
data_setup.py β Download, preprocess, audit, and prepare all datasets for
CodonFM-80M mRNA stability fine-tuning and benchmarking.
Usage:
# Download and preprocess everything
python data_setup.py --all
# Download only training datasets
python data_setup.py --training
# Download only benchmark datasets
python data_setup.py --benchmark
# Audit datasets (inspect stats, find issues)
python data_setup.py --audit
# Export preprocessed training data to local files
python data_setup.py --training --export ./processed_data
# Show codon vocabulary and tokenizer details
python data_setup.py --vocab
"""
import argparse
import json
import os
import sys
import csv
import urllib.request
from collections import Counter
from pathlib import Path
import numpy as np
try:
import pandas as pd
except ImportError:
pd = None
try:
from datasets import load_dataset
except ImportError:
load_dataset = None
# ============================================================
# 1. CODON VOCABULARY & TOKENIZER
# ============================================================
RNA_BASES = ['A', 'U', 'G', 'C']
ALL_CODONS = [b1 + b2 + b3 for b1 in RNA_BASES for b2 in RNA_BASES for b3 in RNA_BASES]
# Biological codon table (RNA) β Amino Acid
CODON_TABLE = {
'UUU': 'Phe', 'UUC': 'Phe', 'UUA': 'Leu', 'UUG': 'Leu',
'CUU': 'Leu', 'CUC': 'Leu', 'CUA': 'Leu', 'CUG': 'Leu',
'AUU': 'Ile', 'AUC': 'Ile', 'AUA': 'Ile', 'AUG': 'Met/Start',
'GUU': 'Val', 'GUC': 'Val', 'GUA': 'Val', 'GUG': 'Val',
'UCU': 'Ser', 'UCC': 'Ser', 'UCA': 'Ser', 'UCG': 'Ser',
'CCU': 'Pro', 'CCC': 'Pro', 'CCA': 'Pro', 'CCG': 'Pro',
'ACU': 'Thr', 'ACC': 'Thr', 'ACA': 'Thr', 'ACG': 'Thr',
'GCU': 'Ala', 'GCC': 'Ala', 'GCA': 'Ala', 'GCG': 'Ala',
'UAU': 'Tyr', 'UAC': 'Tyr', 'UAA': 'Stop', 'UAG': 'Stop',
'CAU': 'His', 'CAC': 'His', 'CAA': 'Gln', 'CAG': 'Gln',
'AAU': 'Asn', 'AAC': 'Asn', 'AAA': 'Lys', 'AAG': 'Lys',
'GAU': 'Asp', 'GAC': 'Asp', 'GAA': 'Glu', 'GAG': 'Glu',
'UGU': 'Cys', 'UGC': 'Cys', 'UGA': 'Stop', 'UGG': 'Trp',
'CGU': 'Arg', 'CGC': 'Arg', 'CGA': 'Arg', 'CGG': 'Arg',
'AGU': 'Ser', 'AGC': 'Ser', 'AGA': 'Arg', 'AGG': 'Arg',
'GGU': 'Gly', 'GGC': 'Gly', 'GGA': 'Gly', 'GGG': 'Gly',
}
# Token vocabulary (matches CodonFM config: vocab_size=69, pad_token_id=3)
SPECIAL_TOKENS = {'[CLS]': 0, '[SEP]': 1, '[MASK]': 2, '[PAD]': 3, '[UNK]': 4}
CODON_TO_ID = {codon: i + 5 for i, codon in enumerate(ALL_CODONS)}
ID_TO_CODON = {v: k for k, v in CODON_TO_ID.items()}
ID_TO_CODON.update({v: k for k, v in SPECIAL_TOKENS.items()})
VOCAB_SIZE = len(SPECIAL_TOKENS) + len(ALL_CODONS) # 5 + 64 = 69
assert VOCAB_SIZE == 69
PAD_TOKEN_ID = 3
CLS_TOKEN_ID = 0
SEP_TOKEN_ID = 1
MASK_TOKEN_ID = 2
UNK_TOKEN_ID = 4
def seq_to_codons(seq: str) -> list:
"""Split an mRNA/DNA sequence into codon triplets."""
seq = seq.upper().replace('T', 'U').strip()
return [seq[i:i+3] for i in range(0, len(seq) - len(seq) % 3, 3)]
def tokenize_mRNA(seq: str, max_length: int = 2046) -> dict:
"""Tokenize an mRNA/DNA sequence into CodonFM token IDs."""
codons = seq_to_codons(seq)
token_ids = [CLS_TOKEN_ID]
for codon in codons[:max_length - 2]:
token_ids.append(CODON_TO_ID.get(codon, UNK_TOKEN_ID))
token_ids.append(SEP_TOKEN_ID)
attention_mask = [1] * len(token_ids)
return {'input_ids': token_ids, 'attention_mask': attention_mask}
def validate_sequence(seq: str) -> dict:
"""Validate an mRNA/DNA sequence for CodonFM compatibility."""
seq_clean = seq.upper().replace('T', 'U').strip()
issues = []
if len(seq_clean) == 0:
issues.append("Empty sequence")
if len(seq_clean) % 3 != 0:
issues.append(f"Length {len(seq_clean)} not divisible by 3 (truncated to {len(seq_clean) - len(seq_clean) % 3})")
invalid_chars = set(seq_clean) - {'A', 'U', 'G', 'C'}
if invalid_chars:
issues.append(f"Invalid characters: {invalid_chars}")
codons = seq_to_codons(seq_clean)
n_codons = len(codons)
# Check for start codon
starts_with_aug = codons[0] == 'AUG' if codons else False
# Check for stop codons
stop_codons = {'UAA', 'UAG', 'UGA'}
internal_stops = [i for i, c in enumerate(codons[:-1]) if c in stop_codons]
ends_with_stop = codons[-1] in stop_codons if codons else False
if internal_stops:
issues.append(f"Internal stop codons at positions: {internal_stops}")
# Unknown codons
unk_codons = [c for c in codons if c not in CODON_TO_ID]
if unk_codons:
issues.append(f"Unknown codons: {set(unk_codons)}")
return {
'valid': len(issues) == 0,
'issues': issues,
'length_nt': len(seq_clean),
'length_codons': n_codons,
'starts_with_AUG': starts_with_aug,
'ends_with_stop': ends_with_stop,
'n_internal_stops': len(internal_stops),
}
# ============================================================
# 2. TRAINING DATASETS
# ============================================================
TRAINING_DATASETS = {
'mogam-ai/CDS-BART-mRNA-stability': {
'description': 'iCodon vertebrate mRNA stability profiles (human, mouse, frog, fish)',
'source_paper': 'Diez et al. 2022, Scientific Reports "iCodon customizes gene expression based on the codon composition"',
'seq_col': 'seq',
'label_col': 'y',
'splits': {'train': 'train', 'val': 'val', 'test': 'test'},
'label_meaning': 'mRNA half-life z-score (higher = more stable, meanβ0, stdβ1)',
'species': ['Human', 'Mouse', 'Xenopus (frog)', 'Zebrafish'],
'notes': 'RNA sequences (A,U,G,C). All divisible by 3. Subset of GleghornLab dataset.',
},
'GleghornLab/mrna_stability_other': {
'description': 'Extended multi-species mRNA stability data (superset of mogam-ai dataset)',
'source_paper': 'Li et al. 2024, Genome Research "CodonBERT large language model for mRNA vaccines"',
'seq_col': 'rna',
'label_col': 'labels',
'splits': {'train': 'train', 'val': 'valid', 'test': 'test'},
'label_meaning': 'mRNA half-life z-score (higher = more stable)',
'species': ['Multiple vertebrate species'],
'notes': 'Has extra "seqs" column (protein-encoded, not used). Contains 1 outlier sequence of 3 nt. Superset of mogam-ai.',
'extra_col': 'seqs',
},
}
def download_training_data(export_dir=None):
"""Download and inspect training datasets from HuggingFace Hub."""
if load_dataset is None:
print("ERROR: `datasets` library required. Run: pip install datasets")
return None
all_data = {}
for repo_id, info in TRAINING_DATASETS.items():
print(f"\n{'='*60}")
print(f"π¦ {repo_id}")
print(f" {info['description']}")
print(f"={'='*60}")
ds = load_dataset(repo_id)
for split_name, hf_split in info['splits'].items():
split_data = ds[hf_split]
seqs = split_data[info['seq_col']]
labels = split_data[info['label_col']]
print(f"\n [{split_name}] {len(seqs)} samples")
print(f" Seq lengths (nt): min={min(len(s) for s in seqs)}, "
f"mean={np.mean([len(s) for s in seqs]):.0f}, "
f"max={max(len(s) for s in seqs)}")
print(f" Seq lengths (cod): min={min(len(s)//3 for s in seqs)}, "
f"mean={np.mean([len(s)//3 for s in seqs]):.0f}, "
f"max={max(len(s)//3 for s in seqs)}")
labels_arr = np.array(labels)
print(f" Labels: min={labels_arr.min():.3f}, mean={labels_arr.mean():.3f}, "
f"std={labels_arr.std():.3f}, max={labels_arr.max():.3f}")
all_data[repo_id] = ds
if export_dir:
export_training_data(all_data, export_dir)
return all_data
def preprocess_training_data(use_both_datasets=True, min_codons=3, max_codons=2046,
remove_duplicates=True, deduplicate_across_splits=True):
"""
Preprocess training data: clean, filter, deduplicate, and combine.
Steps:
1. Load both HF datasets
2. Use GleghornLab as primary (superset) OR combine both
3. Filter: remove sequences < min_codons or > max_codons codons
4. Filter: remove sequences with NaN labels
5. Filter: remove sequences with invalid characters
6. Deduplicate: remove exact sequence duplicates within each split
7. Deduplicate: ensure no train sequences appear in val/test (data leakage check)
8. Return clean {train, val, test} dictionaries
Returns:
dict with 'train', 'val', 'test' keys, each containing 'sequences' and 'labels' lists
"""
if load_dataset is None:
raise ImportError("datasets library required: pip install datasets")
print("Loading datasets...")
if use_both_datasets:
# Use GleghornLab (superset) β it contains ALL of mogam-ai plus extra data
ds = load_dataset("GleghornLab/mrna_stability_other")
raw_data = {
'train': {'seqs': ds['train']['rna'], 'labels': ds['train']['labels']},
'val': {'seqs': ds['valid']['rna'], 'labels': ds['valid']['labels']},
'test': {'seqs': ds['test']['rna'], 'labels': ds['test']['labels']},
}
print(f" Using GleghornLab/mrna_stability_other (superset)")
else:
# Use mogam-ai only (smaller, cleaner)
ds = load_dataset("mogam-ai/CDS-BART-mRNA-stability")
raw_data = {
'train': {'seqs': ds['train']['seq'], 'labels': ds['train']['y']},
'val': {'seqs': ds['val']['seq'], 'labels': ds['val']['y']},
'test': {'seqs': ds['test']['seq'], 'labels': ds['test']['y']},
}
print(f" Using mogam-ai/CDS-BART-mRNA-stability only")
clean_data = {}
total_removed = {'short': 0, 'long': 0, 'nan': 0, 'invalid': 0, 'duplicate': 0}
for split in ['train', 'val', 'test']:
seqs = raw_data[split]['seqs']
labels = raw_data[split]['labels']
orig_count = len(seqs)
clean_seqs = []
clean_labels = []
seen = set()
for seq, label in zip(seqs, labels):
# Skip None/empty
if seq is None or len(seq) == 0:
total_removed['invalid'] += 1
continue
# Normalize: uppercase, TβU
seq = seq.upper().replace('T', 'U').strip()
# Check NaN label
if np.isnan(label):
total_removed['nan'] += 1
continue
# Check invalid characters
if set(seq) - {'A', 'U', 'G', 'C'}:
total_removed['invalid'] += 1
continue
# Check length
n_codons = len(seq) // 3
if n_codons < min_codons:
total_removed['short'] += 1
continue
if n_codons > max_codons:
total_removed['long'] += 1
continue
# Deduplicate within split
if remove_duplicates:
if seq in seen:
total_removed['duplicate'] += 1
continue
seen.add(seq)
clean_seqs.append(seq)
clean_labels.append(float(label))
clean_data[split] = {'sequences': clean_seqs, 'labels': clean_labels}
print(f" [{split}] {orig_count} β {len(clean_seqs)} samples "
f"(removed {orig_count - len(clean_seqs)})")
# Cross-split deduplication: check for train/test leakage
if deduplicate_across_splits:
test_seqs = set(clean_data['test']['sequences'])
val_seqs = set(clean_data['val']['sequences'])
leakage_test = sum(1 for s in clean_data['train']['sequences'] if s in test_seqs)
leakage_val = sum(1 for s in clean_data['train']['sequences'] if s in val_seqs)
val_test_overlap = len(val_seqs & test_seqs)
print(f"\n Data leakage check:")
print(f" TrainβTest overlap: {leakage_test} sequences")
print(f" TrainβVal overlap: {leakage_val} sequences")
print(f" ValβTest overlap: {val_test_overlap} sequences")
if leakage_test > 0 or leakage_val > 0:
print(f" β οΈ WARNING: Data leakage detected! Removing leaked sequences from train...")
eval_seqs = test_seqs | val_seqs
filtered_train = [(s, l) for s, l in
zip(clean_data['train']['sequences'], clean_data['train']['labels'])
if s not in eval_seqs]
clean_data['train']['sequences'] = [x[0] for x in filtered_train]
clean_data['train']['labels'] = [x[1] for x in filtered_train]
print(f" Train after dedup: {len(clean_data['train']['sequences'])} samples")
print(f"\n Removal summary: {total_removed}")
print(f" Final sizes: train={len(clean_data['train']['sequences'])}, "
f"val={len(clean_data['val']['sequences'])}, "
f"test={len(clean_data['test']['sequences'])}")
return clean_data
def export_training_data(data, export_dir):
"""Export preprocessed data to CSV files."""
os.makedirs(export_dir, exist_ok=True)
if isinstance(data, dict) and 'train' in data and 'sequences' in data.get('train', {}):
# Already preprocessed format
for split in ['train', 'val', 'test']:
if split not in data:
continue
filepath = os.path.join(export_dir, f'{split}.csv')
with open(filepath, 'w', newline='') as f:
writer = csv.writer(f)
writer.writerow(['sequence', 'stability_score'])
for seq, label in zip(data[split]['sequences'], data[split]['labels']):
writer.writerow([seq, label])
print(f" Exported {split}: {len(data[split]['sequences'])} rows β {filepath}")
else:
print(" Export requires preprocessed data. Run preprocess_training_data() first.")
# ============================================================
# 3. BENCHMARK DATASETS
# ============================================================
CODONBERT_BASE_URL = "https://raw.githubusercontent.com/Sanofi-Public/CodonBERT/master/benchmarks/CodonBERT/data/fine-tune"
BENCHMARK_DATASETS = {
'stability': {
'url': f"{CODONBERT_BASE_URL}/mRNA_Stability.csv",
'filename': 'mRNA_Stability.csv',
'description': 'mRNA Stability (iCodon vertebrate mRNA half-life)',
'source': 'Diez et al. 2022, Scientific Reports',
'samples': 65356,
'seq_length': '3-3066 nt (1-1022 codons)',
'label': 'Half-life z-score (continuous, meanβ0, stdβ1)',
'metric': 'Spearman Ο',
'columns': 'sequence, value, dataset, split',
'species': 'Multi-vertebrate (human, mouse, frog, fish)',
},
'mrfp': {
'url': f"{CODONBERT_BASE_URL}/mRFP_Expression.csv",
'filename': 'mRFP_Expression.csv',
'description': 'mRFP Protein Expression in E. coli',
'source': 'Li et al. 2024, Genome Research (CodonBERT)',
'samples': 1459,
'seq_length': '678 nt (226 codons, fixed)',
'label': 'Fluorescence intensity (log scale, range 7.4-11.4)',
'metric': 'Spearman Ο',
'columns': 'sequence, value, dataset, split',
'species': 'E. coli (synthetic mRFP variants)',
},
'vaccine': {
'url': f"{CODONBERT_BASE_URL}/CoV_Vaccine_Degradation.csv",
'filename': 'CoV_Vaccine_Degradation.csv',
'description': 'SARS-CoV-2 mRNA Vaccine Degradation',
'source': 'CodonBERT benchmark (derived from Stanford OpenVaccine)',
'samples': 2400,
'seq_length': '81 nt (27 codons, fixed)',
'label': 'Degradation score (z-normalized, range -7.2 to 6.5)',
'metric': 'Spearman Ο',
'columns': 'sequence, value, dataset, split',
'species': 'Synthetic SARS-CoV-2 mRNA vaccine fragments',
},
'riboswitch': {
'url': f"{CODONBERT_BASE_URL}/Tc-Riboswitches.csv",
'filename': 'Tc-Riboswitches.csv',
'description': 'Tetracycline Riboswitch Activity',
'source': 'Groher et al. 2018 (via CodonBERT)',
'samples': 355,
'seq_length': '66-75 nt (22-25 codons)',
'label': 'Switching factor (continuous, range -0.3 to 3.1)',
'metric': 'Spearman Ο',
'columns': 'sequence, value, dataset, split',
'species': 'Synthetic tetracycline riboswitches',
},
'mlos': {
'url': f"{CODONBERT_BASE_URL}/MLOS.csv",
'filename': 'MLOS.csv',
'description': 'MLOS Flu Vaccine Antigen Expression',
'source': 'Ren et al. 2024 (HELM/MLOS)',
'samples': 167,
'seq_length': '~1700 nt (~567 codons)',
'label': 'Expression level (continuous, range 0.3-2.2)',
'metric': 'Spearman Ο',
'columns': 'cds, value (no split column β uses random 70/15/15)',
'species': 'Influenza haemagglutinin CDS variants',
'notes': 'No pre-defined splits. Column name is "cds" not "sequence".',
},
}
def download_benchmark_data(data_dir='./benchmark_data'):
"""Download all benchmark datasets."""
os.makedirs(data_dir, exist_ok=True)
for task_name, info in BENCHMARK_DATASETS.items():
filepath = os.path.join(data_dir, info['filename'])
if os.path.exists(filepath):
size = os.path.getsize(filepath)
print(f" β {info['filename']} already exists ({size/1024:.1f} KB)")
else:
print(f" β Downloading {info['filename']}...")
try:
urllib.request.urlretrieve(info['url'], filepath)
size = os.path.getsize(filepath)
print(f" β Downloaded {info['filename']} ({size/1024:.1f} KB)")
except Exception as e:
print(f" β Failed to download {info['filename']}: {e}")
return data_dir
# ============================================================
# 4. AUDIT
# ============================================================
def audit_dataset(sequences, labels, name="dataset"):
"""Run a comprehensive audit on a list of sequences and labels."""
print(f"\n{'='*60}")
print(f"AUDIT: {name} ({len(sequences)} sequences)")
print(f"{'='*60}")
if len(sequences) == 0:
print(" (empty)")
return
# ---- Sequence stats ----
lengths_nt = [len(s) for s in sequences]
lengths_codon = [len(s) // 3 for s in sequences]
print(f"\n π Sequence Lengths:")
print(f" Nucleotides: min={min(lengths_nt)}, mean={np.mean(lengths_nt):.0f}, "
f"median={np.median(lengths_nt):.0f}, max={max(lengths_nt)}")
print(f" Codons: min={min(lengths_codon)}, mean={np.mean(lengths_codon):.0f}, "
f"median={np.median(lengths_codon):.0f}, max={max(lengths_codon)}")
# Length distribution buckets
buckets = [0, 100, 300, 500, 1000, 2000, 3000, 10000]
hist = np.histogram(lengths_codon, bins=buckets)[0]
print(f" Codon length distribution:")
for i, count in enumerate(hist):
pct = 100 * count / len(sequences)
bar = 'β' * int(pct / 2)
print(f" {buckets[i]:>5}-{buckets[i+1]:>5} codons: {count:>6} ({pct:>5.1f}%) {bar}")
# Sequences > 2046 codons (CodonFM max)
over_limit = sum(1 for c in lengths_codon if c > 2046)
if over_limit > 0:
print(f" β οΈ {over_limit} sequences exceed CodonFM max (2046 codons) β will be truncated")
# ---- Nucleotide composition ----
all_chars = Counter()
for s in sequences:
all_chars.update(s.upper())
total_bases = sum(all_chars.values())
print(f"\n 𧬠Nucleotide Composition:")
for base in ['A', 'U', 'G', 'C']:
count = all_chars.get(base, 0)
pct = 100 * count / total_bases
print(f" {base}: {count:>12,} ({pct:.1f}%)")
unexpected = {k: v for k, v in all_chars.items() if k not in 'AUGC'}
if unexpected:
print(f" β οΈ Unexpected characters: {unexpected}")
# Not divisible by 3
not_div3 = sum(1 for s in sequences if len(s) % 3 != 0)
if not_div3 > 0:
print(f" β οΈ {not_div3} sequences not divisible by 3")
# ---- Codon usage ----
codon_counts = Counter()
for s in sequences[:5000]: # sample for speed
codons = seq_to_codons(s)
codon_counts.update(codons)
print(f"\n π€ Codon Usage (top 10 / bottom 10 from {min(5000, len(sequences))} seqs):")
sorted_codons = codon_counts.most_common()
for codon, count in sorted_codons[:10]:
aa = CODON_TABLE.get(codon, '?')
print(f" {codon} ({aa:>9s}): {count:>8,}")
print(f" ...")
for codon, count in sorted_codons[-10:]:
aa = CODON_TABLE.get(codon, '?')
print(f" {codon} ({aa:>9s}): {count:>8,}")
# Start/stop codon analysis
starts_with_aug = sum(1 for s in sequences if s[:3].upper().replace('T', 'U') == 'AUG')
stop_codons = {'UAA', 'UAG', 'UGA'}
ends_with_stop = sum(1 for s in sequences
if seq_to_codons(s)[-1] in stop_codons) if sequences else 0
print(f"\n π¦ Start/Stop Codons:")
print(f" Starts with AUG: {starts_with_aug}/{len(sequences)} ({100*starts_with_aug/len(sequences):.1f}%)")
print(f" Ends with stop: {ends_with_stop}/{len(sequences)} ({100*ends_with_stop/len(sequences):.1f}%)")
# ---- Label stats ----
labels_arr = np.array(labels, dtype=float)
nan_count = np.isnan(labels_arr).sum()
labels_clean = labels_arr[~np.isnan(labels_arr)]
print(f"\n π Label Distribution:")
print(f" Count: {len(labels_arr)}, NaN: {nan_count}")
if len(labels_clean) > 0:
print(f" Min: {labels_clean.min():.4f}")
print(f" Q1: {np.percentile(labels_clean, 25):.4f}")
print(f" Median: {np.median(labels_clean):.4f}")
print(f" Q3: {np.percentile(labels_clean, 75):.4f}")
print(f" Max: {labels_clean.max():.4f}")
print(f" Mean: {labels_clean.mean():.4f}")
print(f" Std: {labels_clean.std():.4f}")
print(f" Skew: {float(((labels_clean - labels_clean.mean()) ** 3).mean() / labels_clean.std() ** 3):.4f}")
# ---- Duplicates ----
unique_seqs = len(set(sequences))
dup_count = len(sequences) - unique_seqs
print(f"\n π Duplicates:")
print(f" Unique sequences: {unique_seqs}")
print(f" Duplicate sequences: {dup_count}")
# ---- Outliers ----
if len(labels_clean) > 0:
q1, q3 = np.percentile(labels_clean, [25, 75])
iqr = q3 - q1
lower = q1 - 3 * iqr
upper = q3 + 3 * iqr
outliers = np.sum((labels_clean < lower) | (labels_clean > upper))
print(f"\n β‘ Outliers (>3 IQR):")
print(f" Label outliers: {outliers}/{len(labels_clean)}")
very_short = sum(1 for c in lengths_codon if c < 10)
very_long = sum(1 for c in lengths_codon if c > 1000)
print(f" Very short (<10 codons): {very_short}")
print(f" Very long (>1000 codons): {very_long}")
def run_full_audit():
"""Run audit on all training and benchmark datasets."""
print("=" * 70)
print("FULL DATASET AUDIT")
print("=" * 70)
# Training datasets
print("\n\nπ TRAINING DATASETS")
print("=" * 70)
if load_dataset is not None:
ds1 = load_dataset("mogam-ai/CDS-BART-mRNA-stability")
for split in ['train', 'val', 'test']:
audit_dataset(
ds1[split]['seq'], ds1[split]['y'],
f"mogam-ai/CDS-BART-mRNA-stability [{split}]"
)
ds2 = load_dataset("GleghornLab/mrna_stability_other")
for split, hf_split in [('train', 'train'), ('val', 'valid'), ('test', 'test')]:
audit_dataset(
ds2[hf_split]['rna'], ds2[hf_split]['labels'],
f"GleghornLab/mrna_stability_other [{split}]"
)
# Cross-dataset analysis
print("\n\nπ CROSS-DATASET ANALYSIS")
print("=" * 60)
ds1_all = set(ds1['train']['seq']) | set(ds1['val']['seq']) | set(ds1['test']['seq'])
ds2_all = set(ds2['train']['rna']) | set(ds2['valid']['rna']) | set(ds2['test']['rna'])
print(f" mogam-ai total unique: {len(ds1_all)}")
print(f" GleghornLab total unique: {len(ds2_all)}")
print(f" Overlap: {len(ds1_all & ds2_all)}")
print(f" mogam-ai β GleghornLab: {ds1_all.issubset(ds2_all)}")
print(f" GleghornLab-only: {len(ds2_all - ds1_all)}")
else:
print(" Skipping (datasets library not installed)")
# Benchmark datasets
print("\n\nπ BENCHMARK DATASETS")
print("=" * 70)
if pd is not None:
data_dir = download_benchmark_data()
for task_name, info in BENCHMARK_DATASETS.items():
filepath = os.path.join(data_dir, info['filename'])
if not os.path.exists(filepath):
continue
df = pd.read_csv(filepath)
df.columns = [c.lower().strip() for c in df.columns]
seq_col = 'sequence' if 'sequence' in df.columns else 'cds'
val_col = 'value'
if seq_col in df.columns and val_col in df.columns:
audit_dataset(
df[seq_col].tolist(), df[val_col].tolist(),
f"Benchmark: {task_name} ({info['filename']})"
)
else:
print(" Skipping (pandas not installed)")
# ============================================================
# 5. VOCAB DISPLAY
# ============================================================
def show_vocab():
"""Display the full codon vocabulary with amino acid mapping."""
print("=" * 70)
print("CodonFM Tokenizer Vocabulary (vocab_size=69)")
print("=" * 70)
print("\n SPECIAL TOKENS:")
print(f" {'ID':>4} {'Token':<10} {'Description'}")
print(f" {'β'*4} {'β'*10} {'β'*30}")
descriptions = {
'[CLS]': 'Classification token (prepended)',
'[SEP]': 'Separator token (appended)',
'[MASK]': 'Mask token (for MLM pretraining)',
'[PAD]': 'Padding token (pad_token_id=3)',
'[UNK]': 'Unknown token (for invalid codons)',
}
for token, tid in sorted(SPECIAL_TOKENS.items(), key=lambda x: x[1]):
print(f" {tid:>4} {token:<10} {descriptions.get(token, '')}")
print(f"\n CODON TOKENS (64 codons β 20 amino acids + 3 stop):")
print(f" {'ID':>4} {'Codon':<6} {'Amino Acid':<12} {'ID':>4} {'Codon':<6} {'Amino Acid':<12} "
f"{'ID':>4} {'Codon':<6} {'Amino Acid':<12} {'ID':>4} {'Codon':<6} {'Amino Acid'}")
print(f" {'β'*4} {'β'*6} {'β'*12} {'β'*4} {'β'*6} {'β'*12} "
f"{'β'*4} {'β'*6} {'β'*12} {'β'*4} {'β'*6} {'β'*12}")
items = sorted(CODON_TO_ID.items(), key=lambda x: x[1])
for i in range(0, len(items), 4):
row_parts = []
for j in range(4):
if i + j < len(items):
codon, tid = items[i + j]
aa = CODON_TABLE.get(codon, '?')
row_parts.append(f" {tid:>4} {codon:<6} {aa:<12}")
else:
row_parts.append(f" {'':>4} {'':6} {'':12}")
print("".join(row_parts))
print(f"\n TOKENIZATION EXAMPLE:")
example = "AUGGCAGCCGAGACUCGG"
codons = seq_to_codons(example)
tokens = tokenize_mRNA(example)
print(f" Input: {example}")
print(f" Codons: {' '.join(codons)}")
print(f" Token IDs: {tokens['input_ids']}")
decoded = [ID_TO_CODON.get(t, '?') for t in tokens['input_ids']]
print(f" Decoded: {' '.join(decoded)}")
print(f"\n CONFIG (matches nvidia/NV-CodonFM-Encodon-80M-v1/config.json):")
print(f" vocab_size: 69")
print(f" pad_token_id: 3 ([PAD])")
print(f" max_position_embeddings: 2046 codons (~6138 nt)")
print(f" position_embedding_type: rotary (RoPE, ΞΈ=10000)")
# ============================================================
# CLI
# ============================================================
def main():
parser = argparse.ArgumentParser(
description="Dataset setup for CodonFM-80M mRNA stability fine-tuning",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
Examples:
python data_setup.py --all # Download & audit everything
python data_setup.py --training # Download training datasets
python data_setup.py --training --export ./data # Download & export to CSV
python data_setup.py --benchmark # Download benchmark datasets
python data_setup.py --audit # Full audit of all datasets
python data_setup.py --vocab # Show codon vocabulary
python data_setup.py --preprocess # Preprocess & deduplicate
"""
)
parser.add_argument('--all', action='store_true', help='Download and audit everything')
parser.add_argument('--training', action='store_true', help='Download training datasets from HF Hub')
parser.add_argument('--benchmark', action='store_true', help='Download benchmark datasets from GitHub')
parser.add_argument('--audit', action='store_true', help='Run full dataset audit')
parser.add_argument('--preprocess', action='store_true', help='Preprocess training data (clean, deduplicate)')
parser.add_argument('--vocab', action='store_true', help='Show codon vocabulary and tokenizer')
parser.add_argument('--export', type=str, default=None, help='Export directory for preprocessed CSVs')
parser.add_argument('--benchmark_dir', type=str, default='./benchmark_data',
help='Directory for benchmark data')
parser.add_argument('--use_both', action='store_true', default=True,
help='Use both training datasets (default: True)')
parser.add_argument('--mogam_only', action='store_true',
help='Use only mogam-ai dataset (smaller, cleaner)')
args = parser.parse_args()
# Default: show help
if not any([args.all, args.training, args.benchmark, args.audit, args.preprocess, args.vocab]):
parser.print_help()
return
if args.vocab or args.all:
show_vocab()
if args.training or args.all:
print("\n\nπ¦ DOWNLOADING TRAINING DATASETS")
print("=" * 60)
download_training_data(export_dir=args.export)
if args.benchmark or args.all:
print("\n\nπ¦ DOWNLOADING BENCHMARK DATASETS")
print("=" * 60)
download_benchmark_data(args.benchmark_dir)
if args.preprocess or args.all:
print("\n\nπ§ PREPROCESSING TRAINING DATA")
print("=" * 60)
use_both = not args.mogam_only
clean_data = preprocess_training_data(use_both_datasets=use_both)
if args.export:
export_training_data(clean_data, args.export)
if args.audit or args.all:
print("\n\nπ RUNNING FULL AUDIT")
run_full_audit()
if __name__ == '__main__':
main()
|