Developmental DTU patterns

A systematic framework for detecting transient developmental transcript-usage patterns in the developing mouse brain

Read the manuscript · Journal-formatted article · Supplementary figures · Download a release

---

Overview

This repository accompanies a study of transient differential transcript usage (DTU) during mouse brain development.

The framework turns stage-specific DTU evidence into bounded diverge–reconverge episodes. It is model-agnostic and operates after the upstream statistical analysis.

Headline result: 1,348 candidate episodes across 735 genes, centred on an E15.5 midbrain discontinuity.

Original thesis figure: Shtn1 methylation and expression across development.

Study at a glance

Component	Result
Starting set	12,517 isoforms from 4,577 multi-isoform genes
Primary scan	1,348 candidate episodes across 735 genes
Dependence-robust sensitivity	852 episodes across 474 genes
Ranked reciprocal panel	Scg3, Gpm6a, Ntrk2, Tecr, Armc8, Bin1
Secondary methylation audit	11,002 methylation–isoform tests
Prospective methylation candidates	Gnao1 and Taok3, with measurement-sensitive support

The candidates are hypotheses for independent testing. The analysis does not claim that methylation causes the observed transcript changes.

How the framework works

Step	Decision
1	Start with stage-specific DTU evidence from an upstream model
2	Check effect direction, size, and statistical support
3	Require agreement across the comparison groups
4	Confirm reconvergence at the immediate temporal flanks
5	Check separation using biological replicates
6	Retain bounded candidate episodes
7	Rank genes deterministically from the retained evidence

The decision layer requires ordered stages, biological replication, one focal group, and at least two comparison groups.

Repository guide

Path	Contents
manuscript.pdf	Comprehensive 40-page archival manuscript
submission_main.pdf	33-page journal-facing main article
supplementary_figures.pdf	Separate 14-page supplementary file
dtu_analysis/	DTU framework, scripts, derived data, figures, and audit tables
methylation_analysis/	Methylation association and robustness analyses
tables/	Machine-readable claim and specification tables
figures/	Audited manuscript graphics
journal_upload_figures/	Journal-sized figure derivatives
RELEASE_CONTENTS.tsv	File sizes and SHA-256 hashes for the release
CITATION.cff	Citation metadata for GitHub and reference managers

Companion R package

transientDTU implements the reusable post-inference decision layer.

Version 0.99.1:

• accepts generic upstream pairwise-DTU evidence;
• detects bounded diverge–reconverge episodes;
• checks replicate separation;
• annotates reciprocal events;
• produces deterministic gene rankings.

Its installed regression recipe reproduces all 1,348 archived episodes and the ordered six-gene panel.

Reproduce the manuscript

Clone the repository:

git clone https://github.com/Kohze/developmental-dtu-patterns.git
cd developmental-dtu-patterns

Build the comprehensive manuscript:

powershell -NoProfile -ExecutionPolicy Bypass -File .\build.ps1

Build the journal-facing article and supplementary PDF:

powershell -NoProfile -ExecutionPolicy Bypass -File .\build_submission.ps1

The build scripts use a local MiKTeX installation when no explicit -TexBin is supplied. The large local MiKTeX toolchain is not stored in Git.

Analysis scripts and environment records are documented in dtu_analysis/ and methylation_analysis/. Third-party raw inputs are not redistributed.

Release and provenance details

The release builder selects files through fail-closed allow-lists. It combines the manuscript with both analysis namespaces and writes a deterministic ZIP.

python .\release_package.py --config .\release_config.json --destination C:\path\to\combined-paper-release

The repository includes:

• a mixed MIT and CC BY 4.0 licence;
• explicit source-rights confirmations;
• a machine-readable claim ledger;
• figure and text provenance registers;
• a SHA-256 release manifest.

Raw reads, third-party source datasets, archived .RData and .rds objects, local caches, and superseded drafts are excluded.

See PUBLIC_RELEASE_MANIFEST.md and RELEASE_STATUS.md.

Manuscript and figure details

The manuscript contains nine main and 14 supplementary figure environments. It displays 24 graphical assets.

The provenance register covers 32 audited figure PDFs. Journal-upload derivatives preserve vector content and meet the recorded raster-resolution checks.

See figure_provenance.csv and JOURNAL_FIGURE_AUDIT_2026-08-11.md.

Citation

If this repository contributes to your work, cite the companion manuscript:

Gounder R, Hamilton R. A systematic framework for detecting transient developmental transcript-usage patterns identifies an E15.5-centred midbrain candidate landscape. Version 0.1.0, 2026.

Use GitHub’s Cite this repository menu to export the citation from CITATION.cff.

If you use the reusable decision layer, also cite transientDTU 0.99.1.

Licence

• Code and build tooling: MIT
• Manuscript, figures, tables, and author-created derived data: CC BY 4.0

Third-party source material remains under its original terms.