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

This repository accompanies the SEERS manuscript (BioRxiv):
https://www.biorxiv.org/content/10.1101/2025.06.09.658412v2

This repo is focused on two practical parts:

1. SEERS count/enrichment processing and k-mer analysis (R scripts)
   Folder: SEERS_data_process_LiangN/
2. TALE model training/evaluation scripts and a pretrained weight (Python scripts)
   Folder: TALE_models_LiJY_260128/

• Nn_pp.R
  Extracts and counts N45 sequences from merged FASTQ files.
• Nn_pp_pool.R
  Pools counts across biological/technical replicates.
• combine_dna_cyt_nuc.R
  Computes enrichment-style values using DNA/Cytoplasm/Nucleus counts.
• kmer_profiling.R
  Performs k-mer level association/correlation analyses.

• Re_trained_seerr.py
  Main retraining script for SEERS/TALE style modeling.
• eval_seers_lstm_on_3pL6_A549.py
  Evaluation script for LSTM model on A549-related set.
• external_test_3pL6_HCT116.py
  External test script for HCT116-related set.
• eval_external_models_on_3pL6_A549.py
  Compare/evaluate external models on A549-related set.
• eval_clinvar_snps_pytorch.py
  ClinVar SNP effect evaluation utility.
• eval_cnn1_kernel_sweep.py
  CNN kernel sweep evaluation script.
• Randomseq_vs3utr.py
  Utility for random-sequence vs 3′UTR comparisons.
• model/final_model.pth
  Included pretrained model weight.

Use NGmerge as in the manuscript workflow:

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

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

Note: these scripts assume your input paths/filenames are configured inside scripts.

cd TALE_models_LiJY_260128
python Re_trained_seerr.py --train_csv ./train_data_260128/TALE-train-data-260128.csv --save_path ./models/seerr_torch_260128 --seeds 42
python eval_seers_lstm_on_3pL6_A549.py --models_root ./models/seerr_torch_260128 --test_csv ./train_data_260128/3pL6-A549-T1.csv
python external_test_3pL6_HCT116.py
python eval_external_models_on_3pL6_A549.py --test_csv ./train_data_260128/3pL6-A549-T1.csv
python eval_clinvar_snps_pytorch.py
python eval_cnn1_kernel_sweep.py --train_csv ./train_data_260128/TALE-train-data-260128.csv

Full training dataset (Zenodo):
https://doi.org/10.5281/zenodo.18737939

For current TALE scripts, prefer the latest split-aware files from Zenodo:

• TALE-train-data-260128.csv (single file with group column: train / val / test)
• 3pL6-A549-T1.csv (independent high-quality A549 external test set)

The training/evaluation scripts in TALE_models_LiJY_260128/ now support direct use of this group column format for easier standalone reproduction.

• R >= 4.0
• Python >= 3.9
• NGmerge

If you encounter package errors, please install dependencies required by each script in your local environment.