Leaflet-analysis

This repository contains the analysis pipelines and scripts for the results presented in: Aging-associated alternative splicing programs conserved between human and mouse tissues (bioRxiv, 2025).

The core methodology utilizes the LeafletFA model.

• Source Code: LeafletFA Repository
• Utility Functions: LeafletFA-utils

Workflow Overview

Our study's methods includes the following steps:

1. Junction Calling: Raw single-cell BAM files to Splice Junctions (Snakemake).
2. Junction Processing: Filtering and merging across batches.
3. ATSE Mapping: Mapping Alternative Transcript Structure Events (ATSE).
4. Anndata Integration: Aligning splicing data with pre-processed total gene expression.
5. Mouse Foundation Training: Training LeafletFA on the mouse dataset.
6. Cross-Species Comparison: Identifying conserved splice junctions between mouse and human.
7. Transfer Learning: Applying the trained mouse LeafletFA model to human data.

1. Data Processing (BAM to Junctions)

We utilized Snakemake pipelines to process raw data from several major single-cell atlases.

Mouse Datasets

• Allen Brain: LeafletFA_Submission2025/AllenInst/mouse_brain_dev_2021/dataprocessing/snakemake/run_snakemake.sh
• Tabula Muris Senis: (Processed separately by age: 3m, 18m, 24m) LeafletFA_Submission2025/TabulaSenis/raw_data_processing/SS2_Snakemake

Human Datasets

• Allen Brain (Lein Cortex): LeafletFA_Submission2025/AllenInst/lein-cortex-gru/snakemake/run_snakemake.sh
• Tabula Sapiens: LeafletFA_Submission2025/tabula_sapien/snakemake/snakemake/run_snakemake.sh

2. ATSE Mapping and Filtering

After generating junction calls, we combine single-cell data to map Alternative Transcript Structure Events (ATSE) across data sources for each species.

Processing Steps

1. Batching: Junction files are split into 100 batches for parallel processing.
2. Junction Filtering: Using the JunctionReader module from LeafletFA-utils.

• min_intron: 50 bp
• max_intron: 500,000 bp
• min_cells: 2
• min_reads: 10

3. Final ATSE Mapping: Merged junctions must meet a global threshold of 100 reads and 10 cells.
4. Annotation: Each junction is validated against a long-read based GTF, FASTA, and a GenomeDB to ensure valid splice site motifs and annotate exon-exon boundaries.

Scripts

• Mouse: LeafletFA_Submission2025/Mouse_Splicing_Foundation/ATSE_mapper
• Human: LeafletFA_Submission2025/Human_Splicing_Foundation/ATSE_mapper

3. Anndata Preparation and Alignment

Mapped ATSEs and junction counts are integrated into Anndata objects and aligned with gene expression count objects, including scVI normalized expression values and latent space representations.

Note: lncRNAs were excluded from this study.

Species	Alignment Scripts
Mouse	LeafletFA_Submission2025/Mouse_Splicing_Foundation/GeneExpression/01_prepare_expression_data.py to 06_marker_gene_sanity_check.ipynb
Human	LeafletFA_Submission2025/Human_Splicing_Foundation/GeneExpression/01_load_raw_datasets.ipynb to 09_marker_gene_sanity_check.ipynb

4. Mouse Model Training

The LeafletFA model is initially trained on the Mouse Splicing Foundation dataset.

• Training Workflow: LeafletFA_Submission2025/Mouse_Splicing_Foundation/model_train/MOUSE_FOUNDATION/full_workflow

5. Cross-Species Analysis and Transfer Learning

After establishing the mouse foundation, we compare splicing events across species and perform transfer learning.

Conserved Junction Identification

To obtain the list of conserved splice junctions between mouse and human:

• Script: LeafletFA_Submission2025/Multi_Species_Splicing_Foundation/joint_model_train/00_compare_ATSEs.ipynb

Human Transfer Learning

We apply the learned features from the mouse model onto the human dataset using transfer learning:

• Script: LeafletFA_Submission2025/Multi_Species_Splicing_Foundation/joint_model_train/02_run_LeafletFA_transfer.ipynb

Data Availability

Processed data files, including ATSE mapping outputs and Anndata objects used in this analysis, are available on Zenodo:

DOI: 10.5281/zenodo.18158125

Citation

If you use this code or these results in your research, please cite:

Isaev, K. and Knowles, D. A. (2025). Aging-associated alternative splicing programs conserved between human and mouse tissues. bioRxiv.

Contact and Support

If you have any questions, please direct them to karin.isaev[@]gmail.com or submit an issue.