StrainRefine

StrainRefine is a method for refining strain-level metagenomic classification by analyzing read–reference mapping profiles.
It reduces false-positive strain detections by grouping highly similar reference genomes and consolidating their read support.

The method can be used as a standalone post-mapping refinement step or integrated into the MADRe pipeline as a replacement for the original ReadClassification step.

---

Overview

Strain-level metagenomic classification is challenging because reads originating from a single strain often map equally well to several closely related reference genomes. This can lead to redundant detections and large numbers of false-positive strain calls.

StrainRefine addresses this problem by analyzing read–reference mapping profiles:

• constructing binary read-support profiles for candidate reference genomes
• measuring similarity between references based on shared mapped reads
• clustering highly similar references within species
• filtering weakly supported references
• reassigning reads to representative genomes

This consolidation prevents multiple near-identical genomes from being reported separately and helps recover the underlying strain signal even when many highly similar genomes are present in the reference database.

---

Integration with MADRe

StrainRefine was originally developed as a refinement step within the MADRe pipeline.

MADRe consists of two main stages:

1. Database reduction using metagenomic assembly and contig-to-reference mapping
2. Read classification against the reduced reference database

StrainRefine is designed to replace the original MADRe ReadClassification step.
Instead of assigning reads directly to candidate reference genomes, StrainRefine analyzes read–reference mapping profiles to identify groups of highly similar references and refine read assignments.

Although developed for MADRe, StrainRefine can also be used as a standalone post-mapping refinement step for pipelines that produce read–reference mappings in PAF format.

---

Requirements

• Python 3.8+
• Python packages:
  • numpy
  • scikit-learn

Install dependencies using:

pip install numpy scikit-learn

Input

StrainRefine requires the following inputs.

1. PAF file

Read-to-reference mappings in PAF format (e.g. produced by minimap2).

Example: --paf_path mappings.paf

2. Strain–species mapping

JSON file mapping strain taxids to species taxids. The file can be found in database/

Example: --strain_species_info taxids_species.json

---

Usage

Basic usage:

python StrainRefine.py \
    --paf_path mappings.paf \
    --strain_species_info taxids_species.json

Parameters

Required arguments

Parameter	Description
--paf_path	Path to the PAF file containing read–reference mappings.
--strain_species_info	JSON file mapping strain taxids to species taxids. If using the MADRe database, use MADRe/database/taxids_species.json.

---

Clustering

Parameter	Description	Default
--cluster_eps	DBSCAN epsilon value used for clustering references within species.	0.8

---

Reference support thresholds

Parameter	Description	Default
--high_score_threshold	Score threshold used to count high-confidence assigned reads per reference.	0.6
--min_high_score_reads	Minimum number of high-score reads required for a reference to be considered supported.	5
--mid_high_score_reads	Intermediate support threshold for references.	10
--min_high_score_fraction	Minimum fraction of high-score reads required for moderately supported references.	0.8

---

Species-level thresholds (currently informative)

Parameter	Description	Default
--species_min_read_count	Minimum read count threshold for species reliability.	5
--species_min_mean_score	Minimum mean score threshold for species reliability.	0.6
--species_low_count_cap	Species with mean score below threshold and read count below this cap are treated as unreliable.	30

---

Output

Parameter	Description	Default
--read_class_output	Path to the output file containing read classification labels.	read_classification.out
--clustering_out	Directory for clustering-related output files.	clustering_output/
--ref_summary_output	Path to the TSV file containing per-reference read counts and horizontal coverage.	reference_summary.tsv

Output

StrainRefine produces refined read classifications, clustering information, and a per-reference summary.

Read classification

File	Description
read_classification.out	Final strain-level classification labels for reads after refinement.

Each line contains a single read and its assigned reference genome:

read_id : reference_name

---

Reference summary

File	Description
reference_summary.tsv	Per-reference summary with read count and horizontal coverage, sorted by read count in descending order.

This is a tab-separated file with a header line. Each subsequent line corresponds to one final reference genome:

reference    read_count    horizontal_coverage
|511145|NC_000913.3   4821    0.923400
|83334|NC_002695.2     312     0.187200

• reference — reference genome identifier as it appears in the PAF file
• read_count — number of reads assigned to this reference after all refinement and reassignment steps
• horizontal_coverage — fraction of the reference genome length covered by at least one mapped read (0.0–1.0)

---

Clustering output

Directory	Description
clustering_output/	Directory containing clustering results for reference genomes.

The directory contains two files:

File	Description
clusters.txt	Each line represents one cluster and contains reference genome names belonging to that cluster, separated by spaces.
representatives.txt	Each line contains the representative reference genome selected for the corresponding cluster.

The line order corresponds between the two files, meaning that the representative listed on line i in representatives.txt corresponds to the cluster listed on line i in clusters.txt.