Amplicon Coverage Inspector (ACI)

Amplicon Coverage Inspector (ACI) is a bioinformatics tool designed to quantify coverage depth in amplicon-based sequencing data (e.g., ARTIC, Midnight, Swift protocols).

Unlike standard coverage tools that calculate per-base depth, ACI evaluates coverage per-amplicon, providing critical quality control for tiling schemes where overlapping amplicons can create ambiguous read assignments. It is designed to be scheme-agnostic, relying on user-provided BED coordinates rather than rigid primer schemes.

Installation

From PyPI

pip install amplicon_coverage_inspector

From github

git clone https://github.com/erinyoung/ACI.git
cd ACI
pip install .

From bioconda

conda install -c bioconda aci

Dependencies

• python3.9+
  • pandas
  • matplotlib
  • pysam
  • intervaltree

Usage

# Single Sample
aci --bam input.bam --bed amplicon.bed --out results_dir

# Multiple Samples (Explicit list)
aci --bam sample1.bam sample2.bam --bed amplicon.bed --out results_dir

# Batch Processing (Wildcards)
aci --bam *.bam --bed amplicon.bed --out results_dir

Arguments

Argument	Description
-b, --bam	Input BAM file(s). Supports multiple files or wildcards.
-d, --bed	Amplicon BED file (4 columns, no header).
-o, --out	Output directory for results (default: aci).
-t, --threads	Number of threads for sorting and parallel counting (default: 4).
--tmpdir	Custom directory for temporary files (default: system tmp).
--fail-threshold	Min depth required for an amplicon to pass (default: 10).
--fail-percentage	% of samples that must fail for an amplicon to be flagged (default: 50).
-v, --version	Print version and exit.

Methodology: Max vs. Min Logic

In tiling amplicon schemes, amplicons often overlap by ~50bp. A read aligned to this junction physically overlaps two amplicons, creating ambiguity in assignment. ACI quantifies this by calculating two distinct depth metrics:

1. Max Depth (Lenient / Overlap)

• Definition: Counts any read pair that is strictly contained within an amplicon's boundaries.
• Ambiguity: If a read pair fits in the overlap of Amplicon A and Amplicon B, it contributes to the Max Depth of both.
• Interpretation: Represents the theoretical maximum coverage available for the region.

2. Min Depth (Strict / Unique)

• Definition: A subset of Max Depth. Counts a read pair only if it is strictly contained in exactly one amplicon.
• Ambiguity: Reads located in overlapping regions are discarded from this metric.
• Interpretation: Represents the number of fragments uniquely confirmed to originate from a specific amplicon.

Significance: A large divergence between Max and Min depth indicates that the majority of coverage is derived from overlapping regions (or that fragment lengths exceed the non-overlapping amplicon window), which may reduce confidence in variant allele frequency (VAF) calculations for specific amplicons.

Output Files

ACI generates detailed data tables and visualizations in the specified output directory.

Data Tables & Reports

File	Description
amplicon_max_depth.csv	Lenient Depth: Counts reads strictly contained in an amplicon (including those overlapping neighbors).
amplicon_min_depth.csv	Strict Depth: Counts reads strictly contained in only one amplicon (unique assignment).
dropout_report.tsv	QC Report: List of amplicons failing (<10x depth) in >50% of samples.
sample_uniformity_report.tsv	QC Report: Gini coefficient scores indicating coverage evenness per sample.
amplicon_efficiency_matrix.csv	Table showing the % of total reads consumed by each amplicon.
amplicon_read_assignments.csv	Debug: A log of every read ID and which amplicon(s) it was assigned to.

Visualizations

File	Description	Example
amplicon_max_depth_boxplot.png	Boxplot of Max (Lenient) coverage distribution across the cohort.
amplicon_min_depth_boxplot.png	Boxplot of Min (Strict) coverage distribution across the cohort.
depth_comparison_{sample}.png	Per-sample line graph showing the gap between Max and Min depth (overlap uncertainty).
heatmap_max_depth.png	Key Plot: Log-scale heatmap of Lenient Depth.
heatmap_min_depth.png	Key Plot: Log-scale heatmap of Strict Depth. Dark spots indicate dropout.
heatmap_efficiency.png	Linear-scale heatmap showing % of reads per amplicon (detects PCR jackpots).

Input Requirements

1. Pre-processing

Primers should be trimmed from the BAM file prior to analysis. Primer sequences force alignment to the reference and mask variants. ACI is designed to analyze biological insert coverage, not primer binding.

2. Bed File Format

ACI requires a standard 4-column BED file defining the Amplicon (insert) coordinates, not the primer binding sites.

Format: Reference Start End Name (Tab-delimited, no header).

• Reference: Must match the BAM header exactly (e.g., MN908947.3).
• Coordinates: 0-based, half-open.
• Do NOT use a primer scheme BED file. A primer BED file lists binding sites (e.g., Left: 30-54, Right: 385-410). ACI requires the resulting amplicon coordinates (e.g., 55-384).

Example:

MN908947.3	54	385	1	1	+
MN908947.3	342	704	2	2	+
MN908947.3	664	1004	3	1	+
MN908947.3	965	1312	4	2	+
MN908947.3	1264	1623	5	1	+
MN908947.3	1595	1942	6	2	+
MN908947.3	1897	2242	7	1	+
MN908947.3	2205	2568	8	2	+
MN908947.3	2529	2880	9	1	+
MN908947.3	2850	3183	10	2	+

Testing

This repository contains test data in the tests/data subdirectory. To verify the installation:

aci -b tests/data/test.bam -d tests/data/test.bed -o testing

The resulting image should look something like the following.

Background & Motivation

Evaluating primer effectiveness for amplicon-based NGS often requires complex workflows dependent on specific primer scheme formats. While tools like samtools ampliconstats exist, they can be rigid regarding input formats and naming conventions, often failing with custom or highly overlapped primer sets.

ACI was developed to provide a flexible, standalone alternative that strictly evaluates coverage geometry, offering granular insight into library performance regardless of the underlying chemistry.

Contributing

Contributions are welcome! Please feel free to open an issue or submit a pull request.

Future Directions

• Support for single reads
• Support for baits

License

This project is licensed under the MIT License.

Acknowledgments

Google Gemini was used to refactor and modernize this codebase, including introducing a src layout, implementing the strict containment logic, and optimizing parallel processing.