scFiLM manuscript

Manuscript for scFiLM — multi-task learning for perturbation modeling in single-cell transcriptomics.

Target: Briefings in Bioinformatics (OUP) — Problem Solving Protocol Active source: bib-article/bib-skeleton.tex Previous iteration: sn-article-template/sn-article.tex (Springer Nature format; content being ported)

This README doubles as the journal reference sheet. Everything below is sourced from the links in Sources — check against the live pages before submission, since OUP revises them.

---

Article type: Problem Solving Protocol

"Development of methods based on comparison, new software, or new pipeline of existing software for solving a specific bioinformatics problem, with real experimental data that must be analyzed, and should demonstrate a new methodological application to a meaningful biological problem."

A tension worth knowing about. The manuscript-preparation page states that BiB "is a review journal and does not publish articles that are pure original research." Taken alone that reads as disqualifying for a new method. In practice it is not: the Problem Solving Protocol category exists precisely for new software/methods validated on real data, and recently published BiB papers in this category are straightforwardly novel-method papers (see Sanity-check papers).

The practical implication for framing: lean into the comparative dimension. The benchmarking of literature models against scFiLM is not padding — it is what makes the paper legible as a Problem Solving Protocol rather than "pure original research." Position the contribution as solving a stated bioinformatics problem and comparing approaches, not as presenting a novel architecture alone.

Word limits by article type

Type	Words
Review	2,000–7,000
Problem Solving Protocol	2,000–5,000
Case study	2,000–5,000
Opinion / Letter to Editor / Software review / Book review	500–1,000

---

Required elements checklist

Element	Requirement	Draft status
Body word count	2,000–5,000	⚠️ ~6,650 — over by ~1,650
Abstract	Single paragraph, unstructured; ~150–200 words in practice	✅ 143 words
Keywords	6, for indexing	✅ 6
Key Points	3–5 brief sentences, boxed, summarising key messages	⚠️ 3 × TODO placeholders
Author biography	~30 words	⚠️ TODO placeholder
ORCID	Required at submission	⚠️ placeholder 0000-0000-0000-0000
Data availability	Mandatory, in endmatter	⚠️ section exists, needs content
Code availability	Expected where feasible	⚠️ section exists, needs content
Conflict of interest	Mandatory declaration	⚠️ section exists, needs content
Funding	Full agency names + grant numbers	⚠️ section exists, needs content

Submission is via Manuscript Central (mc.manuscriptcentral.com/bib). LaTeX is accepted with the OUP template. Proofs must be checked within 48 hours.

---

Structure

The order below is what recently published BiB Problem Solving Protocol papers actually use (checked against three 2026 papers under literature/: bbag346.pdf, bbag461.pdf, gene_chronos.pdf = bbag469.pdf), and matches the current draft:

Title / Authors / Affiliations / Corresponding author
Abstract                      (single paragraph)
Keywords                      (6)
1. Introduction
2. Methods                    ("Materials and methods" is also used)
3. Results
4. Discussion and/or Conclusion   (see naming note below)
-- endmatter --
Key Points                    (boxed, 3–5 sentences)
Acknowledgements
Author contributions          (CRediT taxonomy)
Supplementary material
Conflicts of interest
Funding
Data availability
Code availability
References
Author biography              (~30 words)
Appendices

Notes:

• Methods precede Results. All three sanity-check papers do this; so does the draft.
• Key Points is an endmatter box, not a title-page one. All three sanity-check papers place it right before Acknowledgements, at the end of the body — not after the abstract/keywords, which is where the vendor oup-authoring-template.cls normally renders anything passed to \boxedtext (that macro is hardwired into \maketitle's title-page output, so the draft reproduces the box manually via a \keypointsbox command placed at the correct spot instead of using \boxedtext).
• The preparation page says the Funding section goes before Acknowledgements, but published articles show Acknowledgements first. Production typesetting normalises this — do not spend effort on it.
• Endmatter order (Acknowledgements → Author contributions → Supplementary material → Conflicts of interest → Funding → Data availability → Code availability) is consistent across all three sanity-check papers and is what the draft now uses.
• "Discussion" vs "Conclusion" is not standardized — observed across the three sanity-check papers:
  • bbag346: single merged section, titled "Conclusion" only.
  • gene_chronos/bbag469: single merged section, titled "Discussion" (plural "Discussions" in the running head).
  • bbag461: two separate sections — "Discussion" (detailed interpretation of results, comparison to prior work, limitations) followed by a short "Conclusions" (brief closing synthesis/future outlook). Either a single merged section (any of those titles) or a split Discussion+Conclusion pair is acceptable; BiB doesn't enforce one. The draft uses a single merged \section{Discussion} — keep it merged rather than splitting, since we're already over the word ceiling and splitting wouldn't remove content, only add a second heading.

---

References

• Numbered, square brackets, in citation order. oup-plain.bst (already used by the draft) is the correct numbered style; oup-abbrvnat.bst is the author-year alternative.
• PubMed conventions, abbreviated journal names:

  Attwood T.K. The role of pattern databases in sequence analysis. Brief Bioinform 2000;1:45–59.

• Four or more authors: first three then "et al."
• Papers "in preparation" cannot be cited.
• Datasets are cited in the reference list as [dataset] Authors, Year, Title, Publisher, Identifier, with accession numbers or DOIs.

Figures and tables

• Line drawings / combination art: 600 dpi. Greyscale and photographs: 300 dpi.
• Figure legends must be understandable independently of the main text.
• Charts: two-dimensional, greyscale backgrounds at 25% tint (guidance aimed at print legibility).
• Tables: column headers must specify units; data right- or decimal-aligned; footnotes lettered and used to expand abbreviations.
• Number figures and tables consecutively.

Style and nomenclature

• CSE Style Manual is the house standard.
• SI units.
• Genotypes italicised; phenotypes not italicised.
• Chemical names per Chemical Abstracts; biochemical and enzyme terminology per IUPAC-IUB.
• Manuscript body double-spaced, A4 or US Letter.
• Define abbreviations at first use; the audience is broad bioinformatics, not perturbation-modeling specialists — expand domain jargon (FiLM, OOD, e-distance) on first appearance.

---

Sanity-check papers

Recent BiB papers used to verify that the guidelines match published practice:

• scGenoByte — transformer for cell type annotation (2026, 27(4)). Intro → Method → Results → Conclusion → Conflicts → Funding → Data availability → References. Abstract ~150–170 words, single paragraph; 5 keywords; 5 Key Points. https://academic.oup.com/bib/article/27/4/bbag369/8725691
• scHILL — individual-level immune heterogeneity from scRNA-seq (2026, 27(3)). Intro → Materials and methods → Results → Conclusion and discussion → Acknowledgements → Author contributions → Conflicts → Funding → Data availability → Code availability → References. 6 keywords; 3 Key Points. https://academic.oup.com/bib/article/27/3/bbag286/8703108

Both are novel-method papers, confirming that the "review journal" framing does not exclude this manuscript.

---

Open gaps in the current draft

1. Trim ~1,650 words to reach the 5,000-word ceiling. The Introduction currently carries a long literature-models subsection ported from the Nature-format draft; that content is valuable for the Problem Solving Protocol framing, so prefer trimming Methods/Results prose and pushing detail into appendices and supplementary material over cutting the comparative framing.
2. Write the 3–5 Key Points (currently TODO).
3. Fill endmatter: funding, conflicts, data/code availability, CRediT author contributions, ~30-word bio, real ORCIDs. notes.md has a worked CRediT example and DOME-ML pointers.
4. Verify every figure legend stands alone without the text.

---

Sources

• Manuscript Preparation — the authoritative page: article types, word limits, references, figures, style conventions
• General Instructions — data availability, conflicts of interest, editorial policy
• Author Guidelines hub
• Journal home
• CRediT taxonomy
• DOME-ML recommendations — ML reporting standards for life sciences
• OUP LaTeX template: vendored in oup-template/oup-authoring-template/, documentation at oup-template/oup-authoring-template/doc/oup-authoring-template-doc.pdf

Legacy

The manuscript was originally drafted against Nature Methods (formatting, submission) in sn-article-template/. notes.md still reflects that target — treat its journal-specific formatting guidance as superseded by this file, while its content outline remains useful.