Running SQANTI BUGSI - ConesaLab/SQANTI3 GitHub Wiki

• You pass --bugsi human (or mouse) on the SQANTI3 command line.
• In src/qc_pipeline.py, SQANTI3 sees args.bugsi and calls:

  generate_bugsi_report(bugsi, outputClassPath, args.isoforms)

• generate_bugsi_report() (in src/qc_output.py) builds and executes:

  Rscript /…/utilities/report_qc/BUGSI_report.R \
    <classification.txt> \
    <bugsi_<human|mouse>.gtf> \
    <your_transcript.gtf> \
    <utilities_path>

• Inputs:
  • classification.txt: SQANTI3 classification of each isoform
  • bugsi_<species>.gtf: gold-standard GTF of known BUGSI genes (with ensembl/refseq/gene_name fields)
  • transcript.gtf: your full transcript GTF
  • path to the utilities directory

1. Load libraries: ggplot2, dplyr, rtracklayer, Gviz, rmarkdown, etc.
2. Import data
   • classification_data ← read SQANTI3 TSV
   • bugsi_gtf ← rtracklayer::import() → extract gene‐level table
   • transcript_gtf ← rtracklayer::import()
3. ID-type classification
   • Classify each associated_gene as "ensembl", "refseq", "gene_name", or "unknown" via regex
   • Choose dominant ID; if Ensembl, strip version suffixes from gene_id in transcripts
4. Clean & explode fusion records
   • Drop fusion records, then re-add them with split associated_gene lists
   • Strip transcript/gene version suffixes and dedupe
5. Define benchmarking sets
   • BUGSI_transcripts: isoforms whose associated_gene is in the gold list
   • TP (True Positives): subset of BUGSI_transcripts with subcategory == "reference_match"
   • PTP (Partial TP): FSM/ISM but not RM
   • FP (False Positives): novel categories (NIC, NNC, genic, fusion)
   • FN (False Negatives): gold‐standard genes with no FSM/ISM hit
6. Compute metrics
   • Sensitivity = # unique TP genes / # gold‐standard genes
   • Non-redundant Precision = TP / total BUGSI_transcripts
   • Redundant Precision = (TP + PTP) / total BUGSI_transcripts
   • Positive Detection Rate = # unique (TP+PTP) genes / # gold‐standard genes
   • False Discovery Rate = (total BUGSI_transcripts – TP) / total BUGSI_transcripts
   • False Detection Rate = FP / total BUGSI_transcripts
   • Redundancy = (FSM + ISM) / # unique (TP+PTP) genes
7. Render report
   • Tabulate and round percentages
   • Assign each isoform to "TP", "PTP", "FP", or "Missing" (for FN)
   • Call rmarkdown::render() on SQANTI3_BUGSI_Report.Rmd

• bugsi_style.css and bugsi_script.js accompany the Rmd to style the interactive report.

• <your_prefix>_BUGSI_report.html in your output directory, containing summary tables, bar/pie charts of TP/PTP/FP/FN, and interactive drill‑downs.

In short:
BUGSI cross‑links your SQANTI3 classification against a curated GTF of known single‑isoform genes, segments isoforms into TP/PTP/FP/FN, computes standard metrics, and wraps everything in a self‑contained RMarkdown HTML report.

• Retrieved GTFs from MANE Select (Human), GENCODE (Human/Mouse), and NCBI RefSeq (Human/Mouse).
• Cross-validation: kept only genes with a single, perfectly matching isoform across all sources (splice junctions, TSS, TTS).
• Initial candidates: 1,925 human genes; 2,345 mouse genes.

• Quantified median expression using GTEx (Human) and ENCODE (Mouse) RNA‑seq.
• Tissue-specific sets: ≥ 5 TPM in at least one tissue.
• Universal set: ≥ 1 TPM across every evaluated tissue.
• Integrated housekeeping genes from HRT Atlas v1.0.

• Multi-exon genes:
  • Extracted annotated junction coverages from Recount3.
  • Computed μ = (Σ Cᵢ) / n.
  • Threshold T = α × μ (α = 0.01).
  • Excluded any gene with novel junction coverage Cₙₒᵥₑₗ > T.
  • Used IntroVerse (Human) to remove genes with novel junctions in > 50% of GTEx samples per tissue.
• Single-exon genes:
  • Overlapped coordinates with refTSS; excluded any with alternative TSS evidence.

• Collaborated with GENCODE annotation experts to verify no plausible alternative isoforms.

• Human: 53 BUGSI genes
• Mouse: 37 BUGSI genes
• Tissue‑specific BUGSI gene lists are available at the BUGSI portal (https://bugsi.uv.es).