tRF target prediction

A pipeline for the prediction of 22nt 3' tRNA fragment (3'-tRF) targets in the mouse genome, with a focus on target sites derived from the primer binding site of LTR-retrotransposons. Associated with the bioRxiv preprint 10.64898/2026.01.22.698655v1. An interactive Shiny app for browsing and filtering annotated target sites is available at https://mwpeacey.shinyapps.io/trf-target-explorer/.

To recover target sites in novel LTR-retrotransposon derived transcripts, a transcriptome is assembled from RNA-seq data from the early embryo. Scripts for transcriptome assembly are adapted from Modzelewski et al 2021 (https://epigenome.wustl.edu/TE_Transcript_Assembly/tool.html).

Except for one-off tasks such as index generation and tRF sequence extraction, individual scripts should be executed with the corresponding wrapper under "wrappers". Editing may be necessary for compatibility with your HPC.

3'-tRF target prediction

1. miranda/generate_tRF_fasta.R: generate a fasta of tRF3b sequences from mature tRNA sequences downladed from GtRNAdb (https://gtrnadb.ucsc.edu).
2. miranda/create_sliding_windows.sh: divide the input genome into windows of 10,000 bp, overlapping by 50 bp, to accommodate memory limitations of miRanda.
3. miranda/miranda_genome.sh: run miRanda using either miRNA settings or custom "tRF" settings that unweight the seed.
4. miranda/gather_summaries.sh: gather summary files into one output folder.
5. R_scripts/miranda_to_bed.sh: process summmary files to .csv and .bed output formats.

Transcriptome assembly

1. If necessary, download public RNA-seq data with "wrappers/download_SRA_wrapper.sh". Metadata or URL inputs can be downloaded from https://sra-explorer.info.
2. Remove adapaters, polyA tails, and low quality bases with "wrappers/trim_wrapper.sh".
3. First pass alignment with "wrappers/STAR_wrapper.sh". If necessary, first generate an index with "transcriptome_assembly/generate_STAR_index.sh".
4. Update the STAR index with new splice junctions with "transcriptome_assembly/generate_STAR_index_second_pass.sh". This requires gathering all SJ.out.tab files from the first alignment into a single directory.
5. Second pass alignment with "wrappers/STAR_wrapper.sh", using the updated STAR index from step 4.
6. Stringtie assembly with "wrappers/stringtie_wrapper.sh". Strandedness must be determined manually (e.g. with "check_strandedness", https://github.com/signalbash/how_are_we_stranded_here)
7. Merge stringtie assembly into a single GTF file and filter for strand information with "transcriptome_assembly/stringtie_merge.sh". This requires moving all .gtf files from step 6 into a single directory.

ORF prediction

1. ORF_prediction/find_orfs.sh: find open reading frames in the stringtie-assembled transcriptome.
2. ORF_prediction/blast_orfs.sh: blast predicted open reading frames against the Refseq annotation. Retains only the top scoring hit.

Annotate target prediction

1. R_scripts/annotate_targets.R: annnotate features of predicted target sites, including their location relative to LTRs and GENCODE/StringTie transcripts.
2. R_scripts/permutation_analysis.sh: calculates the Z-score for overlap between predicted target sites and LTRs at different alignment score thresholds.

Interactive target site browser

1. R_scripts/shiny_target_browser.R: Shiny app for querying annotated target sites, backed by DuckDB.
2. R_scripts/run_shiny_target_browser.R: launch the app locally (default: http://127.0.0.1:3838).
3. R_scripts/deploy_shiny_target_browser_shinyapps.R: deploy the app to shinyapps.io.