1. LandOffer
  2. Blog
  3. All articles
  4. Bioinformatics Portfolio: Make a Public-Data Workflow Reproducible and Reviewable

Job search

Bioinformatics Portfolio: Make a Public-Data Workflow Reproducible and Reviewable

A completed public-orchid FASTA QC packet shows fixed provenance, explicit base-count denominators and an actual controlled failure, repair and repeat.

9 min readLandOffer team

A reviewable bioinformatics portfolio identifies the public data, fixes the workflow and parameters, preserves quality-control results and shows what happened when a run failed. Give the reviewer enough information to repeat the computation and understand its limits. A successful command shows that the declared computation completed; it does not establish a valid biological conclusion.

We completed a small local FASTA workflow over public orchid example sequences from Biopython 1.86. The saved packet includes the exact input and license, source commit, original Python implementation, four actual runs and a read-only verifier. It is a standard-library QC exercise, not a Nextflow or nf-core execution, a sequencing study or a clinical analysis. No human or patient data was used.

LandOffer publishes this guide. Compare the work requested in recent bioinformatics and data roles with the methods you can personally demonstrate. This example shows public-data handling, parameter documentation and failure recovery; it does not claim the broader analysis skills required by every bioinformatics position.

An orchid-themed computational notebook keeps a public-data record, sequence printout and parameter notes together. Editorial illustration, not specimen collection.

Choose a small public input you can identify precisely

The input is Doc/examples/ls_orchid.fasta from the official Biopython repository's biopython-186 tag. Biopython's versioned tutorial identifies the file as orchid DNA sequences and demonstrates reading its sequence records. We used the data file, not the Biopython Python package: the local implementation uses only existing Python standard-library modules. Biopython 1.86 sequence-input tutorial

The saved download contains 76,480 bytes. Its tag resolves to commit c809938fda0ba0a90696779539173ea8d881ad8f; an additional download from that exact commit matched the saved tag-version bytes. The manifest preserves both the source URL and the actual SHA-256 rather than relying on a filename or a moving default branch.

Credit the public data separately from your implementation and review work. We did not collect these orchid sequences, assign their original accession labels or validate their taxonomy. Their headers remain intact. Our work is the bounded local QC implementation, parameter-controlled runs and review packet.

The release's license says unmarked files use the Biopython License Agreement; some files are explicitly dual licensed. We retained the full license and notices instead of calling the entire repository BSD licensed. Record the permission basis and preserve required notices when preparing a shareable copy. Do not assume that every public biological dataset has the same terms. Biopython 1.86 license

Our initial consideration of another public source stopped at a browser-check page. We did not solve or bypass it, and no data from that source entered the workflow. Selecting the accessible licensed orchid example kept the actual input and its provenance clear.

State the workflow contract before interpreting PASS

The authored workflow is labeled version 1.0.0, with its exact script hash recorded. That label is not a published release or an upstream certification. The tested runtime was CPython 3.9.6. No packages, containers, workflow engines or cloud services were installed or used, and the local Nextflow lookup returned no executable.

The script reads declared FASTA, checks record structure and identifier uniqueness, measures sequence lengths and counts bases. Its accepted alphabet is ACGTRYSWKMBDHVN. It retains ambiguous bases, performs no imputation and does not trim or filter sequences. The minimum-length parameter determines whether a record passes this demonstration policy; it does not remove the record from the results.

For the baseline, the declared minimum length is 500 bases. That is an author-chosen exercise setting, not a recommended biological threshold. A different research question would need an appropriate justification and a separate specification. Do not choose a permissive value merely to make a real analysis turn green.

FASTA does not contain per-base sequencing quality scores. This workflow therefore cannot report Phred quality, read trimming performance or sequencing reliability. It also performs no alignment, contamination screening, variant calling, expression analysis or experimental validation. Those are separate tasks with their own inputs, assumptions and methods.

Completed asset one: a fixed workflow and source manifest

The first completed artifact is the versioned computation packet. It connects a specific input to the implementation, runtime, parameters and output identities. A reviewer can distinguish the upstream data release from the author's workflow label and see what changed between runs.

Manifest field Completed record What remains outside the claim
Public source Biopython 1.86 orchid FASTA, pinned tag and exact commit No new collection or patient cohort
Input identity 76,480 bytes and full SHA-256; commit download matches A hash does not validate the original biological annotation
Implementation Owned qc_fasta.py, author version 1.0.0 and script SHA Not an nf-core contribution or a general FASTA library
Environment CPython 3.9.6, standard library only No container, HPC or cross-platform test
Parameters Minimum length, accepted alphabet, ambiguity handling and GC denominator No universal biological QC threshold
Run evidence Separate baseline, failed, repaired and repeat directories No hidden replacement of a failed run with success

The input SHA-256 is:

ea19b38ca97622a601f281439c87ba3edad1f060d9bee05bc0c66536acb6ae09

Each run retains qc.json, a per-record TSV, standard output, standard error and an exit-code receipt. The workflow manifest identifies the script and file hashes. Relative file names in the README make the packet navigable without requiring access to the author's home directory or an authenticated service.

For a single local run from the saved workflow directory, the command is:

python3 qc_fasta.py --input ls_orchid.fasta --min-length 500 --output baseline

The saved runner records the actual commands used for all four cases. It also performs a small pinned-source byte comparison before running the owned script. During a frozen review, use the separate read-only verifier rather than rerunning the writer over the preserved receipts.

nf-core's current specifications discuss end-to-end Nextflow workflows, versioning, containers, documentation and tests. Its mandatory requirements apply to workflows contributed to nf-core. This small custom script has not met those requirements merely because it records parameters and hashes; it makes no nf-core approval or execution claim. nf-core pipeline specifications

Read the actual QC result with its denominator

The baseline parsed 94 records and 67,518 bases, with lengths from 572 to 789 bases. All 94 passed the declared minimum of 500 and the accepted-alphabet check. No duplicate identifiers were found. This check concerns identifiers; the workflow did not deduplicate identical sequence content.

The base counts were A: 15,221; C: 16,228; G: 18,319; T: 17,209; and N: 541. The N positions remained in the data and in the length totals. They were counted as ambiguous rather than silently converted, dropped or treated as invalid under the declared alphabet.

For GC percentage, the script explicitly uses canonical bases only:

(G + C) / (A + C + G + T) × 100
34,547 / 66,977 × 100 = 51.580393%

The denominator excludes the 541 N positions. State the denominator so the reviewer can reproduce the percentage; including all 67,518 positions would change it. This is the aggregate composition of the saved sequences, not an average of the 94 per-record percentages and not a scientific conclusion about a population.

The per-record TSV also makes the ambiguity distribution inspectable. Seven records contain N, and the largest N count in one record is 265. Those actual counts do not explain why the positions are unresolved. They identify where a reviewer would inspect the original record and its annotation if that mattered to a later analysis. We neither fill those positions nor discard the seven records, and we do not call the remaining records experimentally verified.

A PASS result means the records met the particular length and alphabet rules implemented here. It does not mean their annotation, collection process or biological interpretation was verified. The report keeps the original headers, descriptive counts and limitations together so the reviewer can inspect that distinction.

A preserved input record stays untouched while a researcher retains both a failed parameter note and its corrected version. Editorial illustration.

Completed asset two: a failed rerun and an inspected repair

The second completed artifact is the QC and repair report. After the baseline, we deliberately changed only --min-length from 500 to 800. The input file and implementation remained unchanged. Because the observed longest sequence is 789 bases, all 94 records violated that demonstration policy.

The actual failed process exited with code 3 and wrote an explicit QC_FAIL message to standard error. Its separately retained JSON and TSV include all 94 records, each with the reason below_min_length; the saved exit-code receipt and error log preserve the failure. The failure was not a corrupted download, a dropped sequence or an invalid nucleotide symbol.

Actual run Minimum length Exit code and result Saved interpretation
Baseline 500 0; 94 pass Meets the declared teaching contract
Controlled failure 800 3; 94 fail Deliberately incompatible parameter; input unchanged
Repaired 500 0; 94 pass Restores the original contract, not the data
Repeat in a fresh process 500 0; 94 pass QC JSON and TSV hashes match baseline and repair

The repaired run changes the minimum back to 500 while preserving the same input and script hashes. It does not edit sequence bases or hide the failure. The baseline, repaired and repeated runs have identical QC JSON and per-record TSV hashes. The failed output differs because its recorded parameter and dispositions differ.

This is a controlled configuration-recovery exercise. It does not justify lowering a real project's QC standard to get a preferred answer. A real change to a research threshold would need a reason, a documented effect and appropriate review. Here, the deliberately changed parameter is known and the original contract is preserved.

The separate verifier does not import the QC implementation. It independently recounts the saved input, checks all recorded run hashes and confirms the actual failed exit and error log. It also rejects two deliberately wrong in-memory summaries: one dropping the N count and one using the wrong GC denominator. Neither control changes the saved files.

Those checks make the result reviewable for this finite file and environment. They do not establish that the parser handles every FASTA variant or that the chosen checks are sufficient for another biological analysis. This record supports a discussion of configuration recovery; research validity remains outside the test.

Present your contribution and its remaining gaps

A supported portfolio description is: “Built a local standard-library FASTA QC workflow over pinned public orchid example data; retained 94 records, documented ambiguity and GC denominators, preserved a controlled failed run and verified repaired-output equality.” That description identifies the data source and the work actually completed.

Avoid replacing it with “Developed a validated genomics pipeline” or “Executed nf-core analysis” when neither happened. The example contains no sequencing experiment, clinical interpretation or workflow-engine execution. If a role requires those skills, keep them as separate evidence gaps rather than changing the project label.

The nearest machine-learning portfolio guide concerns a fixed evaluation and model errors. This project concerns public sequence provenance, descriptive QC and parameter recovery. Both need reproducible records, but neither method establishes the other's domain expertise.

Before an application, review the requested sample format and confirm that its contents are permitted to share. Open the packet at its README and check that the manifest, result table and failed-run record still refer to the files you intend to share. Compare that evidence with the actual responsibilities in LandOffer's recent roles, and describe the bounded computation rather than an unperformed research result.

Download the worked example files

Download the curated educational companion. Read PUBLICATION-README.md after extraction for the public file layout, replay commands and requirements. The listed offline Python checks were rerun from the extracted archive in a disposable directory. The earlier local report and commands describe the original evidence environment; use the companion README for the portable layout. No whole evidence directory, credentials or compiler binaries are included.

Sources and Further Reading

Public sources and local runs reviewed October 23, 2026. This is a small nonhuman public-data computation, with no patient data, wet-lab work or biological/clinical validity claim.