In an age where pathogens continue to evolve and cross boundaries, understanding what makes them virulent—that is, capable of causing disease—has become a critical focus in modern microbiology and genomics. Virulence prediction bridges computational biology, genomics, and machine learning to forecast the pathogenic potential of microbes before they strike.

What Is Virulence?

Virulence refers to the degree of damage a pathogen can inflict on its host. It is determined by a combination of genetic factors—called virulence factors (VFs)—that allow the organism to attach, invade, evade, and harm the host. These include genes coding for toxins, secretion systems, adhesins, and enzymes that disrupt host defenses.

Understanding virulence factors not only helps in deciphering the mechanisms of infection but also provides early warning signs for emerging threats.

Why Predict Virulence?

Traditional virulence studies relied heavily on experimental infection models, which, although accurate, are time-consuming, expensive, and ethically constrained.
Today, the availability of whole-genome sequences and large-scale pathogen databases has paved the way for in silico virulence prediction—a computational approach that can screen thousands of genomes within hours.

This approach enables researchers to:

• Rapidly identify potential high-risk strains.

• Prioritize pathogens for containment, surveillance, or further study.

• Guide vaccine development and drug target discovery.

• Support One Health frameworks, linking animal, human, and environmental health data.

How Is Virulence Predicted?

Virulence prediction combines bioinformatics pipelines with machine learning and comparative genomics. The process generally involves:

1. Genome Annotation: Identifying genes and coding sequences in microbial genomes.

2. Feature Extraction: Comparing sequences with curated databases like VFDB (Virulence Factor Database), PATRIC, or Victors.

3. Pattern Recognition: Using algorithms (e.g., Random Forest, SVM, or deep learning models) to classify genes or strains as virulent or non-virulent based on sequence patterns, motifs, and protein domains.

4. Scoring and Visualization: Assigning a virulence score or confidence level and visualizing it through heatmaps or genome maps.

Tools and Resources for Virulence Prediction

A number of tools and databases make virulence prediction accessible to the scientific community:

• VFanalyzer – For identifying virulence genes based on VFDB.

• PathoFact – Predicts virulence, antimicrobial resistance (AMR), and toxin genes from metagenomic data.

• Pangenome-based models – Identify virulence-associated gene clusters across strains.

• Machine learning models – Use features like GC content, codon usage bias, or protein domains to predict pathogenicity.

Emerging tools now integrate multi-omic data—including transcriptomics, proteomics, and metabolomics—to understand virulence in a systems biology framework.

Applications in the Real World

Virulence prediction has major implications across public health and research sectors:

• Epidemic preparedness: Early identification of virulent strains in outbreak samples.

• AMR surveillance: Linking virulence profiles with antibiotic resistance determinants.

• Environmental monitoring: Predicting pathogenic potential of soil or waterborne microbes.

• Clinical diagnostics: Supporting personalized treatment through pathogen profiling.

For instance, integrating virulence prediction pipelines into national surveillance networks could enable faster risk assessment and response to infectious outbreaks.

The Road Ahead

As machine learning and genomics advance, virulence prediction will evolve from simple gene-based detection to dynamic, context-aware models that account for host–pathogen interactions, environmental signals, and evolutionary adaptation.

Future tools may predict not just if a strain is virulent, but under what conditions it expresses that virulence—bridging the gap between genotype and phenotype.

In Summary

Virulence prediction is redefining how we understand and anticipate infectious diseases. By coupling genomic insights with computational intelligence, researchers can identify potential threats earlier, design smarter interventions, and ultimately, strengthen our preparedness against emerging pathogens.

Yet, for many students, biologists, and even computer scientists, the question is: “Where do I begin?” With so many platforms, books, and tutorials available, it’s easy to feel overwhelmed.

To make it easier, I’ve compiled 10 excellent resources — ranging from beginner-friendly introductions to advanced computational genomics courses. Many of these are freely available, created by pioneers in the field, and widely used in classrooms and research labs worldwide.

Whether you are a complete beginner or looking to strengthen your foundations, these courses will help you build the skills needed to analyze biological data, design workflows, and think computationally about complex biological systems.

1. HarvardX Data Analysis for Genomics by Rafael Irizarry

From the almighty Rafa, this set of online courses (via edX/HarvardX) is a classic starting point for genomic data science and bioinformatics.

2. Applied Computational Genomics – Aaron Quinlan

Aaron Quinlan (creator of bedtools and many other tools) has made his course materials open. A practical, tool-driven genomics introduction.

3. Bioinformatics Algorithms (Coursera + Companion Book)

Find the highly visual video classes on Coursera, backed by the popular Bioinformatics Algorithms book.

4. The Biostar Handbook

Not a course per se, but a hands-on manual by Istvan (founder of Biostars.org) that’s even used in classes at Penn State.

5. Introduction to Bioinformatics and Computational Biology (by Shirley Liu)

A comprehensive introduction from Shirley Liu’s lab (Harvard DFCI). Covers both theory and computational practice.

6. Data Carpentry: Genomics Workshops

Community-driven training workshops that focus on practical, reproducible research. I was honored to serve as curriculum committee chair here.

7. Computational Genomics: Applied Comparative Genomics

From the Schatz Lab — applied comparative genomics with real-world data.

8. Introduction to Computational Biology (Mike Love, creator of DESeq2)

This course bridges statistics, biology, and computation — a solid primer for anyone entering computational biology.

9. MIT Computational Biology (6.047 / 6.878 / HST.507) by Manolis Kellis

Covers genomes, networks, evolution, and health. A deep-dive from MIT’s OpenCourseWare archive.

10. An Introduction to Applied Bioinformatics

An interactive textbook with Python code, designed for practical applied bioinformatics learning.

Bioinformatics is a rapidly growing field at the intersection of biology, computer science, mathematics, and statistics. As data volumes increase, as well as the diversity of data types (genomics, proteomics, metabolomics, imaging, single‑cell data, etc.), the need for robust computational methods, rigorous models, and reproducible tools has never been greater.

A key decision for researchers is: Where should I publish my work? The choice of journal impacts visibility, peer recognition, and long‑term influence of your research. Below I provide a guide to leading journals in bioinformatics, criteria for selecting the journal that best fits your work, and why these considerations matter.

Leading Journals in Bioinformatics

Criteria to Use When Choosing a Journal

• Scope & Audience
• Impact & Visibility
• Review Time & Speed
• Open Access
• Cost / APCs
• Reputation vs Practical Fit
• Reproducibility, Data & Code Sharing Policies
• Indexing & Reach
• Quality of the field
• Accelerating discovery
• Fair access
• Credibility & trust
• Read recent papers in the journal
• Tailor the manuscript
• Check the author guidelines
• Have backup journals ready
• More emphasis on machine learning / AI
• Single‑cell, spatial omics, multimodal data
• Cloud workflows, reproducible pipelines
• Preprints / open peer review
• Alternative metrics (software use, downloads, community adoption)

Selecting where to publish in bioinformatics isn’t just about prestige; it’s about reaching the right audience, ensuring your work is usable, and contributing to the field responsibly.

1️⃣ Data Analysis for the Life Sciences
A fantastic starting point to learn statistics, R programming, and exploratory data analysis in the context of biology. The best part? It’s available free online from HarvardX.

2️⃣ Practical Computing for Biologists
The very first book I picked up when I started learning computational biology. It’s beginner-friendly and focuses on essential computing skills every biologist needs.

3️⃣ A Primer for Computational Biology
An open-access, hands-on introduction to computational biology concepts and coding techniques. Perfect if you want to learn through real examples.

4️⃣ Computational Genomics with R
For those who already know R and want to dive deeper into genome-scale data analysis, from sequence alignment to gene expression.

5️⃣ The Biologist’s Guide to Computing
Bridges the gap between biological problems and computational thinking, making it easier for life scientists to approach programming and data analysis.

6️⃣ Bioinformatics Data Skills
A must-read to sharpen your bioinformatics toolkit — from command-line skills to reproducible research workflows. Ideal once you’ve covered the basics.

7️⃣ Bioinformatics Workbook
A practical tutorial series to help scientists design bioinformatics projects, analyze data, and understand best practices.

8️⃣ Modern Statistics for Modern Biology
An essential guide to modern statistical methods applied to biology, blending theory with hands-on examples in R.

9️⃣ Algorithms on Strings, Trees, and Sequences by Dan Gusfield
A classic reference for anyone wanting to understand the algorithms behind sequence alignment, genome assembly, and biological data structures.

What are SNPs?
SNPs (pronounced "snips") are positions in the genome where individuals differ by a single nucleotide. For example:

Reference: ...A T G C A T G A...
Variant:     ...A T G T A T G A...

Here, the C in the reference genome has been replaced by a T in the variant.

SNPs occur roughly every 300–1,000 bases in the human genome, meaning there are millions of them scattered throughout our DNA. Most SNPs have no effect on health, but some are linked to disease susceptibility, drug response, and other traits.

Why Do We Analyze SNPs?
1. Medical Genetics

Identify disease-associated variants (e.g., BRCA1/2 in breast cancer).

Predict drug response (pharmacogenomics).

Enable precision medicine by tailoring treatments.

2. Population Genetics & Ancestry

Trace human migration and ancestry.

Study genetic diversity within and between populations.

3. Agriculture & Animal Breeding

Select for desirable traits (drought resistance, yield, disease resistance).

Improve breeding efficiency in livestock.

4. Evolutionary Biology

Track natural selection.

Study adaptation in wild populations.

How is SNP Analysis Performed?
SNP analysis can be broadly divided into three steps:

SNP Detection
Genotyping arrays: Chips that test hundreds of thousands of known SNP positions simultaneously. Fast and affordable, widely used in consumer ancestry testing.

Whole-genome or whole-exome sequencing: Can detect known and novel SNPs across the genome.

Targeted sequencing or PCR: For focused analysis of specific regions.

Variant Calling
Sequencing data is aligned to a reference genome. Bioinformatics tools (e.g., GATK, bcftools) identify positions where the sequenced sample differs from the reference.

Annotation and Interpretation
Tools (e.g., SnpEff, VEP) predict the functional impact of SNPs.

Are the SNPs in coding regions? Do they cause amino acid changes? Are they known to be pathogenic?

Databases like dbSNP, ClinVar, and GWAS Catalog provide information on known associations.

Common Tools for SNP Analysis
Alignment: BWA, Bowtie2

Variant Calling: GATK, FreeBayes

Visualization: IGV, UCSC Genome Browser

Annotation: SnpEff, VEP

Statistical Analysis: PLINK, SNPTEST

Challenges in SNP Analysis
False positives/negatives: Sequencing errors, alignment issues.

Population stratification: Confounding in association studies.

Interpretation: Many SNPs have unknown or complex effects.

Researchers address these with rigorous quality control, large datasets, and increasingly sophisticated statistical models.

The Future of SNP Analysis
With advances in sequencing technology and AI-driven analysis, SNP studies are expanding:

Polygenic risk scores predict disease risk based on thousands of SNPs.

Large-scale biobanks (e.g., UK Biobank, All of Us) enable powerful genome-wide association studies (GWAS).

CRISPR and functional assays help validate SNP effects in the lab.

SNP analysis is at the heart of the genomic revolution, promising insights into biology, health, and evolution at unprecedented scale.

Conclusion
From diagnosing rare diseases to designing better crops, SNP analysis is a foundational tool in modern science. As our ability to sequence and interpret genomes improves, so will our understanding of these tiny—but mighty—variations in DNA.

1. What is a P-Value?
Definition: The p-value is the probability of observing a result at least as extreme as the one you got, assuming the null hypothesis is true.

Low p-value (e.g., p < 0.05) → evidence against the null hypothesis.

High p-value → no strong evidence against the null.

Key idea: It tells you how surprising your data is if there’s really no effect.

2. The Multiple Testing Problem
In bioinformatics, genomics, or any large-scale study, you test thousands of hypotheses (e.g., thousands of genes). Even if there’s no real signal, some tests will have p < 0.05 just by chance.

Example:

Testing 10,000 genes

Even if all null, expect ~500 genes with p < 0.05 by chance

This is why we need multiple testing correction.

3. What is FDR (False Discovery Rate)?
Definition: FDR is the expected proportion of false positives among the results you declare significant.

Unlike the family-wise error rate (FWER), which controls for even a single false positive, FDR lets you tolerate some false discoveries to gain power.

Benjamini–Hochberg (BH) procedure is the most popular method to control FDR.

4. What is a q-value (or q-score)?
Definition: The q-value of a test is the minimum FDR at which that test would be called significant.

A p-value tells you how surprising your result is.

A q-value tells you how many of your significant results might be false positives if you call this result significant.

You can think of the q-value as the FDR-adjusted p-value.

5. Example: Step-by-Step
Let’s work through an example with 10 tests.

Test Raw p-value
1 0.001
2 0.004
3 0.010
4 0.020
5 0.030
6 0.040
7 0.050
8 0.060
9 0.070
10 0.080

Goal: Control FDR at 5%.

Step 1: Rank p-values
Rank from lowest to highest:

Rank p-value
1 0.001
2 0.004
3 0.010
4 0.020
5 0.030
6 0.040
7 0.050
8 0.060
9 0.070
10 0.080

Step 2: Apply Benjamini–Hochberg threshold
For each rank i, compute:

BH critical value =i/m*q
BH critical value=m/i*Q
m = 10 tests
Q = 0.05

Rank p-value BH critical value
1 0.001 0.005
2 0.004 0.010
3 0.010 0.015
4 0.020 0.020
5 0.030 0.025
6 0.040 0.030
7 0.050 0.035
8 0.060 0.040
9 0.070 0.045
10 0.080 0.050

Find the largest p-value ≤ its critical value:

p(4) = 0.020 ≤ 0.020 (T)

p(5) = 0.030 > 0.025 (F)

Result: We can declare the top 4 tests significant at FDR 5%.

Step 3: Computing q-values (conceptually)
The q-value for each p-value is roughly the minimum FDR at which it would be significant. Specialized software (e.g., R’s qvalue package) can estimate them.

In our example:

Tests 1–4 would have q-values ≤ 0.05

Tests 5–10 would have q-values > 0.05

The q-value gives you an adjusted p-value that accounts for multiple testing.

6. In Bioinformatics Workflows
You see these all the time:

RNA-seq differential expression → Report p-values, FDR/q-values

ChIP-seq peak calling

Genome-wide association studies (GWAS)

Proteomics, metabolomics

Always check if results are corrected for multiple testing. Reporting raw p-values alone can be misleading.

Summary
Term Meaning Interpretation
p-value Probability under null Small p → evidence against null
FDR False Discovery Rate Expected proportion of false positives among calls
q-value FDR-adjusted p-value Minimum FDR threshold where result is significant

Final Tip
Always correct for multiple testing! Otherwise, your beautiful "significant" results might just be noise.

So, What Is Data Science?
At its core, data science is the interdisciplinary field that extracts knowledge and insights from data using programming, statistics, and domain expertise. In bioinformatics, data science enables us to turn gigabytes of sequence data into biological meaning.

Imagine trying to understand gene regulation in cancer by analyzing thousands of RNA-seq samples, or predicting antibiotic resistance from bacterial genomes—these challenges are not solvable through wet lab experiments alone. They require data-driven thinking.

Data Science Meets Bioinformatics
Bioinformatics is inherently a data science domain. From genomics to systems biology, every field in modern biology relies on data science techniques to:

Clean and process massive datasets

Discover patterns in high-dimensional data

Build predictive models (e.g., for disease classification)

Visualize complex biological networks and trends

Integrate diverse data types (e.g., transcriptomic + epigenomic data)

The Bioinformatics Toolkit
Here’s what data science typically looks like in bioinformatics:

Task Data Science Role
Sequence alignment Efficient algorithms, indexing, parallel processing
Gene expression analysis Statistical modeling (e.g., DESeq2, limma)
Variant calling Data filtering, probabilistic models
Clustering of cells in single-cell data Unsupervised learning
Protein structure prediction Deep learning models (e.g., AlphaFold)
Metagenomics Data integration, classification, dimensionality reduction

Common tools include Python, R, Bioconductor, scikit-learn, Pandas, Seurat, and TensorFlow—often working together in reproducible workflows.

It's Not Just About Coding
A common misconception is that bioinformatics is just programming or scripting. But being a data scientist in bioinformatics also means:

Understanding experimental design

Asking biologically meaningful questions

Choosing the right statistical or machine learning models

Communicating findings effectively (e.g., plots, dashboards, papers)

In other words, data science in bioinformatics is where biology, statistics, and computer science converge.

Why It Matters
The real power of data science in bioinformatics is its ability to scale discovery.

Instead of studying one gene, we can study thousands.

Instead of analyzing one species, we can explore entire ecosystems.

Instead of waiting months for lab results, we can generate hypotheses in days.

From personalized medicine and cancer diagnostics to agricultural genomics and pandemic surveillance, data science is at the heart of the bioinformatics revolution.

Final Thoughts
If you’re a biologist who’s curious about code, or a data enthusiast fascinated by life sciences, bioinformatics is your playground—and data science is your toolkit.

In bioinformatics, data science isn’t just useful. It’s essential.

Before you go down that path, consider this critical biological question:
Are you sequencing human cells—or bacterial contamination?

The Silent Saboteur: Mycoplasma in Cell Cultures

Mycoplasma contamination remains one of the most widespread and underdiagnosed issues in tissue culture work. Studies suggest that 15–35% of cell lines in use may be contaminated, often without visible signs. Unlike other microbial infections, Mycoplasma does not produce cloudiness, odor, or a change in pH. Many researchers won’t detect it unless they specifically test for it.

The consequences, however, are profound. Mycoplasma can significantly alter:

• Host gene expression patterns

• Cell proliferation rates

• Epigenetic profiles and chromatin accessibility

• Cytokine signaling and immune responses

In short, it can skew your results, compromise your biological conclusions, and invalidate weeks or months of research.

A Simple Diagnostic Step: Map Against Mycoplasma Genomes

If you encounter poor alignment rates to the human genome, consider mapping your reads to a Mycoplasma reference genome—or better yet, use a combined human + Mycoplasma reference. There have been cases where over half of all reads, initially assumed to be from human cells, were in fact bacterial in origin. This check is fast, easy, and could save your project.

How Contamination Happens—and Persists

Mycoplasma is small (0.1–0.3 μm), lacks a cell wall, and can pass through standard filters undetected. Common sources include:

• Contaminated reagents (e.g., FBS)

• Infected cell lines obtained from other labs

• Poor aseptic technique or shared equipment

Once present, it spreads quickly between cultures and can persist for months, silently affecting results.

Why Treatment Is Difficult

While antibiotics such as Plasmocin or BM-Cyclin are sometimes used, they often offer only partial resolution and may themselves alter cell behavior. In many cases, the best course of action is to discard the contaminated culture and start with a fresh, verified stock.

Practical Recommendations for Researchers

• Routinely test for Mycoplasma using PCR, qPCR, or fluorescence-based assays

• Incorporate contamination screens into your sequencing QC pipeline

• Use combined reference genomes when mapping ambiguous reads

• Practice strict aseptic technique and monitor all incoming cell lines

• Don’t ignore unexplained data anomalies—they might point to contamination

Closing Thought: Contamination Is a Biological Variable

It’s easy to view poor mapping as a technical issue, but sometimes the problem lies deeper—in the biology itself. Mycoplasma contamination doesn’t just interfere with sequencing; it interferes with science. As a research community, we must treat contamination not as an afterthought, but as a key variable to control.

So next time your reads won’t align, don’t just tune the aligner. Ask if your cells are telling the truth—or if they're hiding something.

What Does DESeq2 Do?
DESeq2 analyzes count data—the number of sequencing reads that map to each gene. It:

Normalizes the data to account for sequencing depth and library size.

Estimates variance (dispersion) for each gene.

Fits a model to compare groups (e.g., control vs treated).

Calculates fold-changes and p-values to determine significance.

Installing DESeq2

You can install DESeq2 via Bioconductor in R:

if (!requireNamespace("BiocManager", quietly = TRUE))
install.packages("BiocManager")
BiocManager::install("DESeq2")

Inputs Needed

A count matrix: genes as rows, samples as columns (raw counts, not normalized).

A sample metadata table (also called colData): defines the condition/group for each sample.

Example:
# Count matrix (rows = genes, columns = samples)
counts <- read.csv("counts.csv", row.names = 1)

# Sample metadata
colData <- data.frame(
row.names = colnames(counts),
condition = c("control", "control", "treated", "treated")
)

DESeq2 Workflow

1. Load the package
library(DESeq2)
2. Create a DESeqDataSet object
dds <- DESeqDataSetFromMatrix(countData = counts,
colData = colData,
design = ~ condition)
3. Run the differential expression analysis
dds <- DESeq(dds)
4. Get the results
res <- results(dds)
head(res)
This gives a table with:

log2FoldChange: how much expression changed

pvalue: statistical significance

padj: adjusted p-value (FDR corrected)

Visualization (Optional but Powerful)

MA Plot
plotMA(res, ylim = c(-2, 2))

Volcano Plot (custom)
library(ggplot2)
res$significant <- res$padj < 0.05
ggplot(res, aes(x=log2FoldChange, y=-log10(padj), color=significant)) +
geom_point() +
theme_minimal()

Heatmap of Top Genes
library(pheatmap)
topgenes <- head(order(res$padj), 20)
vsd <- vst(dds, blind=FALSE)
pheatmap(assay(vsd)[topgenes, ])

Tips for Best Results
Use raw counts (not normalized or TPM/RPKM values).

Have replicates: DESeq2 relies on variance estimates, so at least 3 per group is ideal.

Watch out for batch effects—include them in your design if needed (e.g., ~ batch + condition).

Summary

Step Purpose
DESeqDataSetFromMatrix() Load your data into DESeq2
DESeq() Run the differential expression analysis
results() Extract the output (log fold change, p-values, etc.)
plotMA() / ggplot2 / pheatmap Visualize the results

Final Thoughts
DESeq2 is an essential tool for RNA-seq data analysis. It abstracts away much of the complexity of statistical modeling, while still giving you control when needed. Whether you're a bioinformatician or a wet-lab biologist, DESeq2 offers both ease of use and analytical power.

What is Magic Wormhole?
Magic Wormhole is an open-source command-line tool that allows you to securely send files or text from one computer to another. Developed by Brian Warner, it aims to eliminate the usual hassle of file transfers: setting up SSH servers, dealing with firewall rules, cloud storage uploads, or even worrying about man-in-the-middle attacks.

Using a combination of PAKE (Password-Authenticated Key Exchange) protocols and end-to-end encryption, Magic Wormhole ensures that the only parties who can see your data are you and your recipient.

“It uses PAKE to establish a secure channel between two computers that use the same one-time code.”

How Does It Work?

One user runs a command like wormhole send file.txt.

The tool generates a human-readable, one-time code (like 7-horse-staple).

The other user types wormhole receive and enters the code.

The file is encrypted, transferred directly (or relayed if needed), and decrypted only on the recipient's side.

All of this happens over a secure channel, with no manual key exchange, configuration, or trust in a central authority.

Example Usage
# Sender
wormhole send myfile.pdf
Sending 1.4 MB file named 'myfile.pdf'
Wormhole code is: 7-horse-staple

# Receiver
wormhole receive
Please enter code: 7-horse-staple
Receiving file (1.4 MB) into: myfile.pdf

That’s it! No email attachments, no cloud storage, no FTP setups.

Why Use Magic Wormhole?
End-to-end encrypted transfers using modern cryptography.

Easy to use even for non-technical users.

Cross-platform: Works on Linux, macOS, and Windows.

No servers needed (except for a lightweight transit relay).

Works even behind NAT/firewalls.

It’s particularly ideal for:

Quickly sharing secrets or passwords.

Distributing software packages securely.

Moving files between servers or VMs.

Under the Hood
Magic Wormhole is written in Python and uses:

SPAKE2 for key exchange.

Transit relay and Mailbox server for message delivery.

Twisted framework for asynchronous networking.

The communication process is decentralized and designed to minimize the trust placed in the relay infrastructure. Even if an attacker intercepts the transit server, they cannot decrypt your data.

Installation

You can install it easily with pip:

pip install magic-wormhole

There’s also a Homebrew package for macOS users:

brew install magic-wormhole
Community and Ecosystem
Magic Wormhole is more than just a file transfer tool. It's part of a growing ecosystem that values user-centric cryptography. There are community-maintained libraries for other languages (e.g., Go, Rust), GUI frontends like wormhole-gui, and integration projects for mobile and web use.

Limitations

While Magic Wormhole is elegant and secure, it’s primarily a command-line utility and not designed for high-volume or persistent file sharing. Transfers require both sender and receiver to be online at the same time. And since it’s peer-to-peer, very large files may suffer performance issues.

Conclusion
Magic Wormhole is a breath of fresh air in the complex world of secure communication. It proves that cryptographic security doesn’t need to come with a heavy user experience cost. If you’re looking for a simple, secure, and delightful way to send files or messages, give Magic Wormhole a try.

Explore the documentation: https://magic-wormhole.readthedocs.io