Recently, Anaconda has decided to start charging users for access to the default channel. This change in policy has significant implications for the data science and machine learning community, as Anaconda is a widely used platform for managing Python and R packages.¹²³ Prefix.dev has a nice article summarizing the whole thing.

The Python ecosystem has recently seen a proliferation of attempts to completely rework its package managers. Several new projects have emerged, each aiming to address in the existing package managers. An issue that isn’t really due that “pip and conda are bad”. pip was released April 4th, 2011 and Anaconda was first release July 17th, 2012. That’s long before I even took an intro CS course my freshman year of college, so I’m not gonna judge what everyone else was doing then.

The space is ripe for disruption. There have been some exciting complete rewrites that have come out recently with ruff and uv from Astral in the Python ecosystem. In the conda ecosystem, I’ve lost track of the different ways to install a conda package, Anaconda, miniconda, Mamba, micromamba, mambaforge. I can’t keep up.

I recently started using Pixi from prefix.dev. It’s been really nice. I want to forget about a package manager. Pixi let’s me do that.

Pixi for Bioinformatics

Set up on a server/locally

curl -fsSL https://pixi.sh/install.sh | bash
# source ~/.bashrc
pixi config append default-channels conda-forge --global
pixi config append default-channels bioconda --global

pixi global install -c bioconda nextflow
pixi g i rclone

Just thought I’d add a quick TL;DR here. This is how easy it is to get started. Works on your server, locally on your laptop. Makes me think of some other great bioinformatics tools…

Things every Bioinformatician loves to see

One way to install it just about everywhere

Problem: Which way should you install conda?

• Anaconda
• miniconda
• Mamba
• Micromamba
• mambaforge

Solution:

curl -fsSL https://pixi.sh/install.sh | bash

Environment activation is automatic

Problem: In every new terminal users have to run conda activate applied-genomics.⁴

Solution: Every project has the same commands

• pixi shell
• pixi shell-hook
• pixi run

Tasks are built-in

Problem: snakemake --cores 4

Solution:

[tasks]
run = "snakemake --cores 4"
upload = { cmd = "rclone sync results/ box:THK_LAB_DATA/results/", depends-on = ["run"] }

pixi run upload

Sure, you could have a Makefile, or reach for just. But that’s just one more tool everyone you collaborate with has to learn.

Storing environment details is taken care of for you

Problem: conda install doesn’t update the environment.yml⁵

Solution: pixi add updates the pixi.toml for you.

But wait there’s more! It locks the version to avoid major version bumps automatically!

[dependencies]
python = ">=3.12.5,<4"

In case python4 sneaks up on you.

No built in lock file

Problem: Versions in the environment.yml are only half the battle…⁶

Solution: pixi add updates the pixi.lock file as well behind the scenes!

It’s gotta be difficult to migrate to, right?

With pixi you can import environment.yml files into a pixi project.

pixi init --import environment.yml

This will create a new project with the dependencies from the environment.yml file.