# Registered single-cell extension report v02

## Design lock

- GSE303359: eight libraries, four biological myoblast lines (two FSHD, two healthy), repeated baseline/6 h injury within line.
- GSE290980: eight libraries, four donor-derived organoid lines (two SMA, two control), two technical libraries per line.
- Cells are used for QC, visualization and marker-defined state localization. They are never inferential replicates.
- FSHD uses line-blocked raw-count pseudobulk; SMA technical libraries are aggregated to line and reported directionally without calibrated p-values.

## FSHD result

The leading prespecified module interaction was DUX4_target (effect 0.392, 95% CI -0.834 to 1.618, FDR 0.883). With two lines per genotype this remains directional and requires independent-line replication.

## SMA result

The largest marker-defined composition difference was Motor_neuron_like (SMA minus control -0.339; MIXED). Labels are transcript-marker definitions, not orthogonally validated cell identities.
- KIF5A line-level log2 effect -0.294 (DOWN_ALL_PAIRWISE).

## Candidate consequence

No candidate is promoted to VALIDATED. The new data refine state localization and the next experiment only.

## Limitations

Two biological lines per genotype cannot support calibrated population inference. Batch correction is restricted to PCA visualization/clustering. Raw-count pseudobulk retains the biological unit. Marker-defined states require orthogonal confirmation, and no result is a treatment recommendation.