SplicingDataset format

SplicingDataset is the shared h5ad format consumed by LeafletFA and SpliceVI. It is produced by ATSEmapper from per-cell junction BED files.

Structure

A SplicingDataset is an AnnData object where:

• rows (obs) = cells
• columns (var) = junctions (one column per junction)

Two sparse layers are required:

Layer	Shape	dtype	Contents
cell_by_junction_matrix	cells × junctions	float32, CSR/COO	Read counts per junction per cell
cell_by_cluster_matrix	cells × junctions	float32, CSR/COO	Total ATSE counts per cell (beta-binomial denominator)

CSR is the preferred sparse format. COO is also accepted.

What is an ATSE?

An Alternative Transcript Structure Event (ATSE) is a group of mutually exclusive or competing junctions that together describe a single splicing choice (e.g. a cassette exon skip produces a skipping junction and two inclusion junctions).

Each junction in adata.var belongs to exactly one ATSE. Multiple columns of both matrices map to the same ATSE.

Understanding cell_by_cluster_matrix

cell_by_cluster_matrix[c, j] is the total reads across all junctions in the same ATSE as junction j, for cell c. It is the denominator in the beta-binomial likelihood and ensures PSI estimates are compositionally consistent within each splicing event.

Key invariants:

• junction[c,j] > 0 ⟹ cluster[c,j] > 0 (can't have junction reads without an ATSE total)
• junction[c,j] = 0 does not imply cluster[c,j] = 0 (other junctions in the same ATSE may have reads)
• For all junctions j1, j2 in the same ATSE: cluster[c, j1] == cluster[c, j2]

ATSE 0 (junctions 0,1,2)              ATSE 1 (junctions 3,4)
          j0   j1   j2                       j3   j4
cell 0 [   3    0    1  | cluster=4 ] [   0    2  | cluster=2 ]
cell 1 [   0    2    0  | cluster=2 ] [   1    0  | cluster=1 ]
cell 2 [   1    1    0  | cluster=2 ] [   0    0  | cluster=0 ]

Sparsity

Both matrices are typically 70–95 % zeros for Smart-Seq2 data. This sparsity is expected and handled correctly by the model; do not impute or fill zeros before passing to LeafletFA.

Constructing from scratch

If you are not using ATSEmapper and want to assemble the AnnData manually:

import numpy as np
import scipy.sparse as sp
import anndata as ad

# 5 cells, 6 junctions grouped into 2 ATSEs (3 junctions each)
atse_ids = np.array([0, 0, 0, 1, 1, 1])  # which ATSE each junction belongs to

junc = np.array([
    [3, 0, 1, 0, 2, 0],
    [0, 2, 0, 1, 0, 3],
    [1, 1, 0, 0, 0, 2],
    [0, 0, 4, 2, 1, 0],
    [2, 0, 0, 0, 3, 1],
], dtype=np.float32)

# cluster[c, j] = sum of all junctions in the same ATSE as j, for cell c
cluster = np.zeros_like(junc)
for atse in np.unique(atse_ids):
    cols = atse_ids == atse
    cluster[:, cols] = junc[:, cols].sum(axis=1, keepdims=True)

adata = ad.AnnData(X=sp.csr_matrix(junc))
adata.layers["cell_by_junction_matrix"] = sp.csr_matrix(junc)
adata.layers["cell_by_cluster_matrix"] = sp.csr_matrix(cluster)

adata.obs (one row per cell) and adata.var (one row per junction) can hold any additional metadata. Cell annotations such as cell_type or tissue added to adata.obs are preserved through training and useful for interpreting learned splicing programs.

Zenodo dataset

The mouse splicing atlas used in the paper is available at DOI: 10.5281/zenodo.18158125:

File	Description
model_ready_aligned_splicing_data.h5ad	Raw junction counts ready to pass to LeafletFA (~1 GB)
splice_adata_for_figures_mouse_foundation.h5ad	Trained atlas with fitted SPs and cell activities (~93 MB)