生物信息分析任务轨迹数据集(转录组、基因组)

{
  "task_type": "bioinfo_analysis",
  "domain": "biogene",
  "execution_status": "success",
  "toolchain": "biopython,pyfaidx,gffutils",
  "language": "en",
  "duration_seconds": 1507.564,
  "task": "Spliced CDS translation consistency audit
     
     Audit CDS extraction and translation from the supplied Saccharomyces cerevisiae S288C chromosome I reference and GFF3 annotations. Group CDS parts by their actual parent relations, order parts by transcription strand, account for GFF3 phase and explicitly flagged partial or exceptional features, and translate using the appropriate nuclear genetic code stated in the annotation or a clearly justified default. Independently compare a gffutils/pyfaidx extraction route with a separate GFF-row/Biopython route. Report agreement, non-multiple-of-three lengths, internal stops, absent start/stop codons, and partial/exceptional cases separately; do not change annotations to force agreement or claim every partial feature is defective. Include positive and negative tests for strand order, phase boundaries and an intentionally changed base in a separate test copy. Save the program, per-CDS audit table, tests, and a summary report. The task is reference-annotation validation, not sequence design.
     
     ",
  "steps": [
    {
      "step": 1,
      "tool": "run_commands",
      "command": "",
      "output_excerpt": "3.12.13 (main, Jun 23 2026, 15:44:24) [Clang 22.1.3 ]
         {'biopython': '1.88', 'gffutils': '0.14', 'pyfaidx': '0.9.0.4'}"
    },
    {
      "step": 2,
      "tool": "run_commands",
      "command": "from collections import Counter,defaultdict
         from urllib.parse import unquote
         from pathlib import Path
         import hashlib,json
         p=Path('input/yeast_chrI_NC_001133.9.gff3')
         counts=Counter(); keys=defaultdict(Counter); groups=defaultdict(list); phases=Counter(); flags=[]
         for n,line in enumerate(p.read_text().splitlines(),1):
          if not line or line.startswith('#'): continue
          c=line.split('\\t'); a={k:[unquote(x) for x in v.split(',')] for k,v in (x.split('=',1) for x in c[8].split(';'))}
          counts[c[2]]+=1
          for k in a: keys[c[2]][k]+=1
          if c[2]=='CDS':
           phases[c[7]]+=1
           for parent in a.get('Parent',['ORPHAN']): groups[parent].append((n,c[3:8],a.get('ID'),a.get('locus_tag')))
          if set(a)&{'exception','transl_except','transl_table','pseudo','pseudogene','ribosomal_slippage'} or (c[2]=='CDS' and 'partial' in a): flags.append((n,c[2],c[3:8],{k:v for k,v in a.items() if k not in ['Note','experiment','Dbxref','product']}))
         print('counts',dict(counts));print('CDS attribute keys',dict(keys['CDS']));print('phases',dict(phases));print('CDS parents',len(groups));print('multipart',json.dumps({k:v for k,v in groups.items() if len(v)>1},indent=2));print('flags',json.dumps(flags,indent=2))
         print('partial by feature type',{k:v['partial'] for k,v in keys.items()});print('hashes',{f.name:hashlib.sha256(f.read_bytes()).hexdigest() for f in Path('input').iterdir() if f.is_file()})
         PY",
      "output_excerpt": "counts {'region': 1, 'telomere': 2, 'origin_of_replication': 11, 'gene': 100, 'mRNA': 95, 'exon': 106, 'CDS': 97, 'long_terminal_repeat': 9, 'regulatory_region': 1, 'ncRNA': 1, 'tRNA': 4, 'snoRNA': 1, 'centromere': 4, 'mobile_genetic_element': 1, 'pseudogene': 1}
         phases {'0': 95, '1': 1, '2': 1}
         CDS parents 94"
    },
    {
      "step": 3,
      "tool": "agent_note",
      "text": "The annotation has 97 CDS rows under 94 direct parents, including three multipart CDSs. All 94 coding transcripts and their genes are flagged partial, but none of the CDS rows are; I’ll report that distinction without treating parent-level partiality as a CDS defect.
         
         No `transl_table` is stated. I’ll use standard nuclear code 1 for S. cerevisiae chromosome I, not the yeast mitochondrial code. For spliced CDSs, internal phase values will be checked at exon boundaries—not used to delete bases that complete a codon across the splice.
         "
    },
    {
      "step": 4,
      "tool": "editor",
      "action": "create audit_cds.py",
      "content_head": "#!/usr/bin/env python3
         \"\"\"Parent-aware, two-route CDS validation; never repairs the input annotation.
         
         Route A: gffutils relations, pyfaidx oriented slices, explicit codon lookup.
         Route B: independent GFF3 rows/Parent graph, SeqIO/Seq slices and translation.
         Shared code is limited to metadata policy, comparison, and reporting.
         \"\"\"",
      "output_excerpt": "File created successfully"
    },
    {
      "step": 5,
      "tool": "run_commands",
      "command": "import audit_cds as a
         x=a.route_gffutils(a.DEFAULT_FASTA,a.DEFAULT_GFF,a.ROOT/'results/work')
         y=a.route_biopython(a.DEFAULT_FASTA,a.DEFAULT_GFF)
         r=a.audit_rows(x,y)
         print(a.summarize(r,a.DEFAULT_GFF)['metrics'])
         print('disagreements',[(v['parent_id'], {k:v[k] for k in v if k.endswith('agree') or k.endswith('agrees')}) for v in r if not v['all_routes_agree']])
         PY",
      "output_excerpt": "{'cds_groups': 94, 'cds_rows': 97, 'multipart_groups': 3, 'plus_strand': 49, 'minus_strand': 45, 'all_routes_agree': 94, 'spliced_nt_agrees': 94, 'translation_agrees': 94, 'raw_non_multiple_of_three': 0, 'coding_non_multiple_of_three': 0, 'groups_with_internal_stops': 0, 'internal_stops': 0, 'absent_start': 0, 'absent_stop': 0, 'canonical_ATG': 94, 'phase_boundary_mismatches': 0, 'checked_internal_boundaries': 3, 'nonzero_initial_phase': 0, 'direct_CDS_partial': 0, 'ancestor_partial': 94, 'partial_context': 94, 'exceptional_CDS': 0, 'overlapping_groups': 0, 'ambiguous_base_groups': 0, 'groups_with_sequence_findings': 0}
         disagreements []"
    },
    {
      "step": 6,
      "tool": "agent_note",
      "text": "The first full comparison agrees for all 94 CDS groups at both nucleotide and amino-acid levels. All have ATG starts and in-frame terminal stops; none has an internal stop, a length remainder, or an inconsistent splice-boundary phase.
         
         I’m now testing those results against wrong strand order, incorrect phase trimming, and a one-base change in a separate reference copy. The changed-base test will also check that agreement between routes does not hide a newly introduced sequence finding.
         "
    },
    {
      "step": 7,
      "tool": "run_commands",
      "command": "chmod +x run.sh; ./run.sh",
      "output_excerpt": "[Command exited with code 1]
         ERROR: test_negative_invalid_structure (test_audit.AuditTests.test_negative_invalid_structure) (case=3)
         AttributeError: 'str' object has no attribute 'decode'. Did you mean: 'encode'?
         Ran 15 tests in 0.307s
         FAILED (errors=1)"
    },
    {
      "step": 8,
      "tool": "editor",
      "action": "edit audit_cds.py: route_gffutils",
      "new_text": "def route_gffutils(fasta, gff, work):
             \"\"\"Use database direct-parent relations, not name or protein-ID heuristics.\"\"\"
             work = Path(work)
             work.mkdir(parents=True, exist_ok=True)
             def validate_feature(feature):
                 # Reject invalid coordinates before gffutils computes SQLite genomic bins.
                 if feature.start is None or feature.end is None or not 1 <= feature.start <= feature.end:
                     raise ValueError(f\"Invalid GFF coordinates on {feature.id}\")
                 return feature
             db = gffutils.create_db(str(gff), dbfn=str(work / \"annotation.sqlite\"),
                                    force=True, keep_order=True, merge_strategy=\"create_unique\",
                                    transform=validate_feature,
                                    disable_infer_genes=True, disable_infer_transcripts=True)"
    },
    {
      "step": 9,
      "tool": "run_commands",
      "command": "./run.sh",
      "output_excerpt": "Ran 15 tests in 0.278s
         OK
         Tests: passed; inputs unchanged: True"
    },
    {
      "step": 10,
      "tool": "run_commands",
      "command": "./run.sh",
      "output_excerpt": "[Command exited with code 1]
         FAIL: test_ambiguous_bases_are_reported_not_hidden (test_audit.AuditTests.test_ambiguous_bases_are_reported_not_hidden)
         AssertionError: 'MBZXA*' != 'MBZJA*'
         Ran 16 tests in 0.259s
         FAILED (failures=1)"
    },
    {
      "step": 11,
      "tool": "editor",
      "action": "edit tests/test_audit.py",
      "old_text": "self.fixture(sequence=\"ATGRAYSAAMTNGCNTAA\", first_length=4)",
      "new_text": "self.fixture(sequence=\"ATGRAYSAAMTHGCNTAA\", first_length=4)"
    },
    {
      "step": 12,
      "tool": "run_commands",
      "command": "./run.sh",
      "output_excerpt": "Ran 16 tests in 0.296s
         OK
         Tests: passed; inputs unchanged: True"
    }
  ],
  "final_result": "Completed the audit and saved all requested outputs.
     
     ### Observed results
     
     | Check | Result |
     |---|---:|
     | CDS rows / parent groups | 97 / 94 |
     | Multipart CDS groups | 3 |
     | Nucleotide and translation agreement | **94/94** |
     | Non-multiple-of-three CDS lengths | 0 |
     | Internal stops | 0 |
     | Absent start / terminal stop codons | 0 / 0 |
     | Inconsistent splice-boundary phases | 0 |
     
     Used **standard nuclear genetic code 1**, justified because this is S. cerevisiae nuclear chromosome I and no `transl_table` is specified.
     
     All 94 coding genes/transcripts carry partial flags, but their CDS rows do not. These flags are reported separately—not interpreted as evidence that the CDSs are defective. The explicitly pseudogenic **YAR061W** has no CDS and is listed separately.
     
     ### Tests and integrity
     
     - **16 tests passed**, including 26 parameterized subtests covering strand order, phase boundaries, partial features, and negative controls.
     - A one-base change in a **separate reference copy** created an internal stop at EFB1 codon 10; the audit detected it.
     - Original inputs remain byte-identical to their supplied checksums.
     - A second complete run produced byte-identical principal outputs.
     
     ### Saved files
     
     - [Summary report](report.md) and [structured report](report.json)
     - [Per-CDS audit table](results/cds_audit.tsv)
     - [Audit program](audit_cds.py)
     - [Tests](tests/test_audit.py) and [test results](results/tests.json)
     - [Rerun script](run.sh)
     - [README and output conventions](README.md)
     
     Recompute from this workspace with:
     
     ```sh
     ./run.sh
     ```
     
     The reports document limitations: route agreement validates extraction/translation consistency, not biological function or agreement with an external protein reference."
}

python

def route_gffutils(fasta, gff, work):
    """Use database direct-parent relations, not name or protein-ID heuristics."""
    work = Path(work)
    work.mkdir(parents=True, exist_ok=True)
    def validate_feature(feature):
        # Reject invalid coordinates before gffutils computes SQLite genomic bins.
        if feature.start is None or feature.end is None or not 1 <= feature.start <= feature.end:
            raise ValueError(f"Invalid GFF coordinates on {feature.id}")
        return feature
    db = gffutils.create_db(str(gff), dbfn=str(work / "annotation.sqlite"),
                           force=True, keep_order=True, merge_strategy="create_unique",
                           transform=validate_feature,
                           disable_infer_genes=True, disable_infer_transcripts=True)