update changelog, pyproject.toml and CITATION.cff

CHANGELOG.md

@@ -5,6 +5,32 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.3.4] - 2026-09-22
+
+### Added
+- `simulate_trial_v2()` and `TranscriptomeSimConfig`: transcriptome-scale hierarchical gamma-Poisson simulator calibrated from real single-cell data, superseding the original `SimulationConfig` API
+- TNBC (triple-negative breast cancer) dataset support and accompanying tutorial
+- Second-order DiD analysis demonstrated in tutorials
+- Signal-fraction sensitivity benchmark (gene panel × signal percentage grid)
+
+
+### Fixed
+- Double log bug in benchmark orchestrator and permutation modules (`orchestrator.py`, `permutation.py`)
+- GSEA ranking coerced to finite, unique values before `gseapy.prerank` — infinite betas from degenerate OLS and duplicate gene symbols caused "float division by zero" across all five AML gene-set libraries
+- Cluster-robust fit falls back to non-robust fit when within-arm degrees of freedom are exhausted, fixing silent failure of AML GSEA
+- NEBULA runner: filter zero-count cells before computing log library-size offset; set `ncore=1` to prevent PSOCK port conflicts; robust R error capture and crash detection
+- COVID loader: CITE-seq antibody-capture features (~49% of aggregate counts) were sitting in the gene matrix, confounding library-size normalisation; split to `obsm["protein"]`
+- Melanoma: sample filter silently dropped 19% of QC-passing cells (3,108 of 16,291) and 7 of 32 patients; per-cell annotations now sourced from the published labels; RECIST arm normalised to leading code
+- Vaccine and AML loaders: data-integrity fixes from 4-team audit
+- Benchmark: signal-count truncation, frozen design leaking into scenarios, split jobs overwriting combined results
+
+### Improved
+- QC: ambient and housekeeping gene classes excluded upstream in all loaders, so GSEA rankings inherit a clean gene set
+- Vaccine loader (GSE171964): raw UMI counts now normalized with 10x QC, matching other loaders
+- Benchmark simulator calibrated from real TNBC data: within-cell-type coherent, mean-dependent dispersion, arm-stratified bootstrap gate
+- 27 mypy type errors fixed across benchmark and dataset modules
+- Tutorials re-executed against `v0.3.4`
+
 ## [0.3.3] - 2026-04-05
 
 ### Added

CITATION.cff

@@ -12,8 +12,8 @@ authors:
   - family-names: "Omar"
     given-names: "Mohamed"
 title: "sctrial: Participant-Level Differential Analysis for Longitudinal Single-Cell Experiments"
-version: 0.3.3
-date-released: 2026-04-05
+version: 0.3.4
+date-released: 2026-09-22
 url: "https://github.com/TheOmarLab/sctrial"
 repository-code: "https://github.com/TheOmarLab/sctrial"
 license: MIT

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "sctrial"
-version = "0.3.3"
+version = "0.3.4"
 description = "Participant-level differential analysis for longitudinal single-cell experiments: DiD, paired comparisons, module scoring, pseudobulk, and power analysis."
 readme = { file = "README.md", content-type = "text/markdown" }
 requires-python = ">=3.9"