CAPHEINE, or everything and the kitchen sink: a workflow for automating selection analyses using HyPhy

CAPHEINE is a portable, open-source computational workflow that automates the end-to-end process of evolutionary selection analysis using HyPhy, transforming raw unaligned pathogen sequences and a reference genome into comprehensive results for site-level, gene-level, and lineage-specific selection studies.

The Gist

Imagine you are a detective trying to solve a mystery: How do viruses change, survive, and sometimes jump from one host to another?

Viruses are like master shapeshifters. They mutate rapidly, creating thousands of slightly different versions of themselves. Some of these changes help them hide from our immune systems, while others might make them better at infecting new animals (like jumping from wild birds to cattle).

The problem is that scientists have so much data—millions of viral genetic sequences—that trying to find the "smoking gun" (the specific mutations that matter) is like trying to find a needle in a haystack while wearing blindfolded.

Enter CAPHEINE.

What is CAPHEINE?

Think of CAPHEINE as an automated, high-tech detective squad for viral evolution. It's a computer program (a workflow) built by researchers at Temple University that does all the heavy lifting for you.

Instead of a scientist spending weeks writing custom code to clean data, align sequences, and run complex statistical tests, they can just feed CAPHEINE two things:

1. The Suspects: A pile of unorganized viral genetic sequences (the "unaligned" data).
2. The Blueprint: A reference genome (the "ideal" version of the virus to compare against).

CAPHEINE then takes over, running a full investigation from start to finish.

How Does It Work? (The Kitchen Sink Analogy)

The authors jokingly call this workflow "everything and the kitchen sink." Here is what happens inside the machine, translated into everyday terms:

1. The Cleanup Crew: First, CAPHEINE sweeps the floor. It removes messy data (like sequences with too many gaps or errors) and aligns all the viral sequences so they line up perfectly, like soldiers in a parade.
2. The Family Tree Builder: It builds a family tree (phylogeny) to show how all these viral strains are related to each other.
3. The Stress Test (Selection Analysis): This is the core magic. CAPHEINE runs six different "stress tests" (using tools called HyPhy) to answer specific questions:
   • FEL & MEME: "Are there specific spots on the virus that are changing too fast?" (This suggests the virus is evolving to escape our defenses).
   • BUSTED: "Is the whole gene under pressure to change?"
   • RELAX & Contrast-FEL: "Is the virus evolving differently in one group compared to another?" (e.g., Is the virus changing faster in cattle than in wild birds?)

The Real-World Test: The H5N1 Bird Flu Mystery

To prove it works, the team used CAPHEINE to investigate the H5N1 bird flu virus. They had a massive dataset:

• The Background: Thousands of strains from wild birds (the virus's natural home).
• The Foreground: Thousands of strains from a recent outbreak in cattle.

They wanted to know: Is the virus changing as it jumps from birds to cows?

The Results:

• The General Rule: Most of the virus stays the same (it's under "purifying selection," meaning nature keeps it stable because changing too much breaks it).
• The Exceptions: CAPHEINE found specific "hotspots" where the virus was changing rapidly.
• The Smoking Gun: They found a specific spot (Site 88) in a gene called M2.
  • In wild birds, this spot usually has one type of amino acid (Aspartate).
  • In cattle, the virus switched to a different amino acid (Asparagine).
  • Why it matters: This gene helps the virus package its genetic material. The change suggests the virus is actively adapting its "packing tape" to work better inside a cow's body.

Why Should You Care?

Before tools like CAPHEINE, doing this analysis required a PhD in computer science just to set up the software. It was slow, prone to errors, and hard to repeat.

CAPHEINE is like a self-driving car for evolutionary biology:

• It's Portable: It runs on Mac, Windows, or Linux (even on supercomputers).
• It's Reproducible: If you run it today and I run it tomorrow, we get the exact same results.
• It's Fast: It turns weeks of work into hours.

The Bottom Line

CAPHEINE doesn't just tell us that a virus is evolving; it tells us where and why. It helps scientists spot the exact mutations that might make a virus more dangerous, better at spreading, or resistant to vaccines.

By automating the boring stuff, CAPHEINE lets researchers focus on the big picture: How do we stop the next pandemic before it starts? It's a powerful new lens that helps us see the invisible battle between viruses and the hosts they infect.

Technical Summary

1. Problem Statement

Evolutionary dynamics in viruses, particularly RNA viruses like Influenza A (IAV), HIV-1, and SARS-CoV-2, are highly diverse and driven by rapid mutation, recombination, and transmission bottlenecks. While vast amounts of viral sequence data are available (e.g., from GISAID and SRA), there is a critical lack of standardized, reproducible, and scalable computational methods to analyze this data for selection signatures.

Existing workflows often suffer from:

• Lack of Portability: Many are tailored to specific pathogens or datasets, making them difficult to adapt.
• Technical Barriers: Generalist tools (e.g., V-pipe, ViralFlow) often focus on single-sample variants and fail to test for selection across lineages or conserved sites.
• Reproducibility Issues: Researchers frequently build custom, non-standardized pipelines for selection analysis, leading to inconsistent results and difficulty in comparing studies.

There is a pressing need for a workflow that is portable across operating systems, supports High-Performance Computing (HPC), and automates the full spectrum of evolutionary selection analyses (site-level, gene-level, and lineage-specific).

2. Methodology

The authors developed CAPHEINE (Comprehensive Automated Pipeline using HyPhy for Evolutionary Inference with Nextflow), a bioinformatics workflow built on the Nextflow language and the nf-core framework.

Workflow Architecture:

• Inputs:
  1. Query Dataset: A single FASTA file of unaligned pathogen genomes (full or partial).
  2. Reference Dataset: A FASTA file of reference coding sequences (e.g., from NCBI RefSeq).
  3. Optional Foreground: A list or regular expression to define specific lineages (foreground) for comparative analysis against the rest (background).
• Preprocessing Steps:
  1. Trimming: Removal of terminal stop codons from reference sequences.
  2. Filtering: Removal of query sequences with >50% gaps or ambiguous bases.
  3. Alignment: Mapping query sequences to reference genes using cawlign (a codon-aware pairwise aligner preserving reading frames).
  4. Deduplication: Removal of duplicate sequences using hyphy cln to speed up computation.
  5. Tree Inference: Construction of Maximum Likelihood trees using IQ-Tree2 (GTR+I+G model).
  6. Branch Labeling: If a foreground set is provided, internal branches are labeled as "Foreground" or "Reference" using HyPhy's label-tree.
• Core Evolutionary Analyses (HyPhy Suite):
  The pipeline executes six specific HyPhy methods:
  1. FEL (Fixed Effects Likelihood): Detects pervasive diversifying selection at each site.
  2. MEME (Mixed Effects Model of Evolution): Detects episodic diversifying selection (selection acting on specific branches).
  3. BUSTED (Branch-Site Unrestricted Statistical Test for Episodic Diversification): Tests for gene-wide evidence of positive selection.
  4. PRIME: Analyzes whether changes in amino acid properties (e.g., hydrophobicity, charge) are favored at specific sites.
  5. Contrast-FEL: Tests if sites evolve differently in the foreground clade compared to the reference.
  6. RELAX: Tests for relaxation or intensification of selection strength in the foreground clade relative to the reference.

Output: Results are generated as plain-text CSV files, facilitating easy integration with R, Python, or spreadsheet software.

3. Key Contributions

• Standardization: CAPHEINE provides a single, reproducible workflow for a suite of popular selection tests, eliminating the need for bespoke scripting.
• Portability & Scalability: Built on Nextflow/nf-core, it supports Docker, Singularity, and Conda, allowing execution on laptops, HPC clusters, and cloud environments without manual environment tuning.
• Comparative Capability: It uniquely integrates Contrast-FEL and RELAX to automatically detect shifts in selection pressure between specific lineages (e.g., host shifts) and the broader population.
• Accessibility: The workflow requires minimal input (unaligned sequences + reference genes) and produces user-friendly outputs, lowering the barrier to entry for evolutionary biologists.

4. Results (Case Study: H5N1 Host Shift)

The authors validated CAPHEINE using H5N1 Influenza A Virus (IAV) genomes (1975–2025), comparing wild bird reservoirs (Anseriformes/Charadriiformes) against a cattle outbreak (2025).

• Dataset: ~28,751 cattle isolates and ~41,988 wild bird isolates.
• General Patterns:
  • Most genes showed strong purifying selection (mean $\omega$ between 0.06 and 0.82).
  • BUSTED analysis revealed episodic diversifying selection in over half of the gene products (PB2, PB1, PA, NA, NS1).
  • PRIME identified sites under selection for various biochemical properties (e.g., secondary structure, charge) across all genes.
• Host Shift Analysis (Cattle vs. Wild Birds):
  • RELAX Analysis: Indicated intensified selection in HA, NS1, and PB2 in cattle, and relaxed selection in NP, PA, and PB1 compared to wild birds.
  • Site-Specific Adaptation:
    • Contrast-FEL and MEME identified two sites with intensified positive selection in cattle: one in PA and one in M2.
    • M2 Site 88: This site showed a shift in the majority amino acid residue (Asparagine in cattle vs. Aspartate in wild birds). Given that M2 residues 82–89 are critical for viral packaging and infectivity, this suggests a functional adaptation to the cattle host.

5. Significance

• Actionable Insights: CAPHEINE transforms raw sequence data into actionable leads for experimental follow-up, vaccine design, and drug target prioritization by pinpointing specific adaptive mutations.
• Public Health Utility: The workflow enables rapid surveillance of emerging variants and host shifts (e.g., zoonotic spillover), providing early warnings of viral adaptation.
• Community Resource: By standardizing the "exploratory" phase of evolutionary analysis, CAPHEINE allows researchers to focus on biological interpretation rather than data wrangling.
• Limitations & Future Work: The authors note that the pipeline does not currently automate recombination screening (essential for HIV/Coronaviruses) and assumes haploid genomes, leaving complex genomic scenarios to the user's discretion.

In summary, CAPHEINE represents a significant step forward in making rigorous, large-scale evolutionary selection analysis accessible, reproducible, and efficient for the broader scientific community.