Skip to main content

Understanding CNV Model Performance Reports

Written by Support

When you create a new CNV model based on a cohort of reference samples, Rainbow generates comprehensive performance reports that summarise the model's quality and reliability. This guide explains each field and metric in these reports to help you interpret the results.

Report Overview

CNV model performance reports consist of two main components:

  1. Summary Report (*_summary.txt): A high-level overview of model performance across all targets, transcripts, and genes

  2. Gene Report (*_gene_report.csv): A detailed per-gene and per-transcript breakdown of quality metrics

Summary Report Fields

METADATA Section

The metadata section provides essential information about when and how the model was generated:

  • Running Date: The date and time when the model generation process was executed

  • Reference Version: The reference genome version used (e.g., HG19, HG38)

  • Model type: whether the model is Exome (omim_panel_exons) or Genome

  • Coverage Stats Type: The type of genomic regions analysed by the model (exons/kit regions)

  • BED File: Name of kit’s BED file

  • Transcript Filter: Which transcripts were included in the report for each gene (All transcripts / canonical transcripts only)

  • Number of Samples: Total count of reference samples in the cohort

PARAMETERS Section

This section documents the configuration settings used during model generation:

  • Kit: The unique model name used

  • Gene Panel Config: Gene panel configured for the model. The default is whole-exome

  • Filtering Options: Additional filtering settings:

    • Filter_samples: Whether correction for common shared variants was applied

    • filter_targets: Limits callable regions to only regions that are included in the model’s BED file

  • Report Thresholds:

    • Score_threshold: minimal model prediction score threshold

    • Percentage_of_region_coverage_threshold and min_coverage_depth: parameters use to detect which exons\targets are considered “low coverage”

STATISTICS Sections (Targets, Transcripts, Genes, General)

The STATISTICS sections summarize model performance at increasing levels of aggregation:

  • Target-level: individual regions/exons

  • Transcript-level: groups targets by transcript

  • Gene-level: groups transcripts/targets by gene

Across these sections, the report lists counts and percentages of items that fall into quality buckets (for example, High / Medium / Low / Failed) and may include additional stats, such as Low Coverage, Invalid Targets, or Problematic Targets.

How to interpret: In general, a strong model has a high fraction of High-quality items and a low fraction of Failed / Low coverage items. If you see elevated failure/low-coverage rates, use the gene-level breakdown (and then the CSV gene report) to pinpoint which genes/transcripts are driving the issue.

Gene Report CSV Fields

The gene report CSV provides detailed per-gene and per-transcript metrics. Each row represents one transcript of a gene.

Basic Identifiers

  • gene: Gene symbol (e.g., "BRCA1", "A1BG")

  • transcript: (in gene reports only) Transcript identifier (e.g., "NM_130786.4")

  • targeted: Boolean indicating whether this transcript has targeted regions in the enrichment kit (true/false)

Region Statistics

  • total_exons: Total number of exons (in gene report) or targets (in kit report) in the gene/transcript that are covered by the model

  • invalid_exons_count: Number of exons that failed quality checks and for which the model will not work

  • low_quality_exons_count: Number of exons classified as low quality, where the model might have lower performance.

  • invalid_exons%: Percentage of exons that failed (invalid_exons_count / total_exons × 100)

  • low_quality_exons%: Percentage of exons that are low quality (low_quality_exons_count / total_exons × 100)

Coverage Metrics

  • avg_coverage_percentage: Between all the targets in this gene, average the percentage of the cohort samples with adequate coverage.

  • avg_median_coverage: the average median coverage of targeted regions in this gene in all the samples of the cohort.

Model Quality Metrics

  • avg_model_score: Average model score across all regions of the gene

  • model_quality: Overall quality classification for this transcript based on the model score:

    • High: All or most exons have high calling quality

    • Medium: Mixed quality with acceptable performance

    • Low: Significant quality concerns

    • Fail: Below calling ability

Quality Indicators

  • failure_reasons: Semicolon-separated list of reasons why invalid exons failed (e.g., "15:LOW_COVERAGE" means exon 15 failed due to low coverage)

  • quality_warnings: Semicolon-separated list of reasons for low-quality exons (e.g., "15:LOW_COVERAGE" means exon 15 has low model quality due to low coverage)

Interpreting the Results

Using the Reports

  1. Review Summary Report First: Get an overall sense of model quality

  2. Check Gene Report for Specific Genes: If analysing specific genes, review their individual metrics

  3. Identify Problematic Regions: Use failure_reasons and quality_warnings to understand issues

  4. Validate Against Expectations: Compare results to your enrichment kit design and expected coverage

Did this answer your question?