A systematic delineation of 3-prime UTR regulatory elements and their contextual associations.

This repository accompanies the SEERS study.

This repository is organized around two main components:

1. SEERS count/enrichment processing and k-mer analysis in SEERS_data_processing/
2. TALE model training, evaluation, and ClinVar ISM analysis in TALE_model_260312/

The materials in TALE_models_260128_legacy/ are retained as a legacy reference.

• Nn_pp.R: extracts and counts N45 sequences from merged FASTQ files.
• Nn_pp_pool.R: pools sequence proportions across replicates.
• combine_dna_cyt_nuc.R: computes cytoplasmic and nuclear enrichment scores from DNA/Cyt/Nuc counts.
• kmer_profiling.R: performs k-mer association analyses.

These scripts use local input filenames configured near the top of each script. Run them from the directory containing the input files, or update work_dir and the input filenames explicitly.

Recommended TALE model files.

• model/model_bundle.pt: selected pretrained PyTorch model bundle.
• model/summary.json: run configuration and metrics.
• notebooks/TALE_training_260312.ipynb: training notebook snapshot.
• notebooks/TALE_ClinVar_ISM_v6.ipynb: ClinVar SNV ISM logo notebook.
• results/: training histories and summary plots.
• SHA256SUMS: checksums for the bundled model and summary.

Selected run:

• TALE_e5-lstm256bx128b-fc256-0.1-ema0.999-seed3407_260320_132204
• Training data: TALE_train_data_260312.csv
• Independent A549 evaluation data: 3pL6-A549-T1.csv
• Selected weights: EMA
• Best validation loss: 0.2024837457580668

Legacy TALE scripts and pretrained weights retained for reference and comparison. For the current TALE model files, use TALE_model_260312/.

TALE training and evaluation data are distributed separately from this code repository.

For the current TALE workflow, use:

• TALE_train_data_260312.csv
• 3pL6-A549-T1.csv

Minimal requirements:

• R >= 4.0
• Python >= 3.9
• NGmerge for paired-end read merging

Python dependencies are listed in requirements.txt.

The optional Python package umap-learn is used only when running UMAP-based legacy k-mer comparison code.

R package dependencies:

• Biostrings
• data.table
• ggplot2
• ggpointdensity
• pbmcapply
• scales
• viridis

Prepare merged FASTQ files when starting from paired-end reads:

./NGmerge -d -1 read1.fq.gz -2 read2.fq.gz -o merged.fq.gz

Run SEERS count and enrichment processing:

cd SEERS_data_processing
Rscript Nn_pp.R
Rscript Nn_pp_pool.R
Rscript combine_dna_cyt_nuc.R
Rscript kmer_profiling.R

Run or inspect the updated TALE notebooks:

cd TALE_model_260312/notebooks

The training notebook expects TALE_train_data_260312.csv and 3pL6-A549-T1.csv to be available locally. The ClinVar ISM notebook uses the bundled model by default and expects ClinVar input sequences to be supplied in the notebook parameter cell.