← Về thư mục
📄 / / profiles / r-and-d / raw_evidence / OpenAI4S / skills / bioskills / bio-copy-number-recurrent-cnv / references / usage-guide.md

Recurrent and Driver CNV Usage Guide

Overview

A copy number change in one tumor is an observation; one recurring across many tumors beyond chance is evidence of selection. GISTIC2 separates recurrent driver events from passengers by modeling a background rate and scoring each locus by frequency and amplitude (the G-score), with separate background estimation for focal and arm-level events. Copy-number signatures decompose the genome-wide alteration pattern into the mutational processes that generated it. This skill covers GISTIC2, driver-gene localization from recurrence peaks, and the Steele 2022 and Drews 2022 copy-number signature frameworks, including the caveats that make their output easy to misread.

Prerequisites

# GISTIC 2.0: download the compiled binary + MATLAB Compiler Runtime from the Broad
# Institute; there is no conda/pip package. Verify the -refgene .mat matches the genome build.
R -e "BiocManager::install('CINSignatureQuantification')"   # Drews 2022 signatures
pip install SigProfilerAssignment                            # Steele 2022 COSMIC CN

Inputs: a pooled cohort segment file (6 columns: sample, chrom, start, end, num_markers, seg.mean), diploid-centered; for signatures, absolute allele-specific copy number from ASCAT/Sequenza/FACETS.

Quick Start

Tell the AI agent what to do: - "Run GISTIC2 on my cohort segment file to find recurrent focal amplifications" - "Localize the driver gene inside this wide GISTIC peak" - "Separate focal drivers from arm-level passenger events in my cohort" - "Quantify copy-number signatures for my tumor cohort" - "Explain why I cannot compare GISTIC q-values between my two cohorts"

Example Prompts

Recurrence and drivers

"Pool these CNVkit segments into a cohort seg file and run GISTIC2 with high peak-boundary confidence, then list focal amplification peaks and their candidate driver genes."

"This GISTIC peak is 3 Mb wide and contains 25 genes. Localize the likely driver using the COSMIC Cancer Gene Census and explain why the peak gene may be a passenger."

Focal vs broad

"Distinguish the focal driver events from arm-level passenger events in my cohort and explain how Ziggurat deconstruction and arm-peeling make that separation."

Signatures

"Quantify copy-number signatures for my cohort with the Drews CINSignatures framework using absolute copy number from ASCAT."

"Explain why three of the Steele 2022 copy-number signatures are oversegmentation artifacts and how that affects my interpretation."

What the Agent Will Do

  1. QC and diploid-center the cohort segmentation before pooling
  2. Run GISTIC2 with focal and broad analysis and high peak-boundary confidence
  3. Localize candidate driver genes from recurrence peaks plus known-driver evidence
  4. Report recurrence frequency (portable) alongside q-values (cohort-size dependent)
  5. Quantify copy-number signatures from absolute CN with one consistent framework
  6. Flag oversegmentation-artifact signatures and caller-sensitivity caveats

Tips