CWI-SSN

🎯 Coverage-Weighted Identity-based Sequence Similarity Network

CWI-SSN is a lightweight pipeline for constructing sequence similarity networks (SSNs) from all-vs-all BLAST results.

Instead of relying on EFI-EST’s alignment score, CWI-SSN uses a biologically intuitive metric — the Coverage-Weighted Identity (CWI) — defined as: $\text{CWI} = \frac{\text{pident}}{100} \times \sqrt{\frac{L_{align}}{L_Q} \times \frac{L_{align}}{L_S}}$

This measure integrates sequence identity with symmetric coverage, providing a length-independent and easily interpretable similarity score (0–1).

CWI-SSN is particularly suited for short or highly variable sequences, such as RiPP precursor/core peptides, where traditional statistical alignment scores lose their significance. Therefore, our analysis focuses on sequence identity rather than bit score or E-value.

🚀 Installation & Quick Start

Create environment and install requirements

mamba create -n modifiedSSN python=3.12 blast=2.16.0 -y
mamba activate modifiedSSN

Install from source

git clone https://github.com/BioGavin/CWI-SSN.git
cd CWI-SSN
pip install .

This will install CWI-SSN with all dependencies and create the cwi_ssn command-line tool.

Note: Always install biopython via conda (mamba install -c bioconda biopython) rather than pip to avoid compilation issues on some systems.

Basic Usage

cwi-ssn -i sequences.faa -o results --cwi-threshold 0.6 --threads 6

Arguments

Argument	Short	Default	Description
--input	-i	required	Protein FASTA file for BLAST database and queries
--output-dir	-o	results	Output directory for BLAST and SSN files
--threads		16	CPU threads for BLASTP
--cwi-threshold		0.6	Minimum CWI score (0-1) for keeping edges
--include-singletons		False	Include sequences without edges as singleton nodes

Example

cwi-ssn -i data/demo.faa -o my_results --cwi-threshold 0.6 --threads 6 --include-singletons

Output Files

The pipeline generates:

• blast/ - BLAST database and all-vs-all results (.m8 format)
• ssn/ - Processed network files:
  • network_cwi*.graphml - Final network in GraphML format
  • Intermediate files (reduced edges, scored edges)

❓ Why not EFI-EST SSN for short peptides

The original EFI-EST framework builds sequence similarity networks (SSNs) using BLAST-based pairwise comparisons, where edges are scored by an alignment score ($AS$) derived from the bit score ($B$) ($AS = -\log_{10}!\left[ 2^{-B} (L_Q L_S) \right]$) . Where, $L_Q$ and $L_S$ are the lengths of the query and subject sequences, respectively.

EFI-EST was originally developed for enzymes of comparable lengths, where a single alignment score threshold reliably correlates with functional similarity. However, for short or length-variable sequences—such as RiPP precursor or core peptides—this assumption fails, resulting in biased similarity scores and fragmented networks. The core issue lies in the length bias of bitscore-derived alignment scores. According to the definition of alignment score, its length dependence originates from the bit score, which in turn derives from the raw alignment score in BLAST. Because the raw alignment score increases proportionally with the length of the aligned region, shorter identical sequences naturally produce lower bit scores and alignment score, relative to longer sequences. Consequently, even sequences that are compositionally identical but differ in length can yield substantially different bit and alignment scores. Therefore, a single alignment scores threshold is not suitable for short or length-variable sequences.

A small illustrative case is documented in the issue file. I have also initiated a discussion regarding this issue at EFI-EST Discussion #222.