ARCade

Background

Why ARC needs a gentler on-ramp

ARC is the University of Calgary's high-performance computing cluster. Getting your first pipeline to run on it means learning a lot at once:

writing job scripts and choosing the right partition, memory and time limit;
installing tools and staging large files the way the cluster expects;
checking on jobs, reading logs, and working out why something failed.

Each step is learnable, but together they can take weeks — and for most research groups, the pipeline is not the science. It's what stands between you and it.

ARCade takes care of that on-ramp. Describe what you want in a chat, and a small AI model — running on a shared ARC server — works with a set of pre-tested pipelines to do the rest. It's built and maintained by the Bioinformatics Hub at the Snyder Institute, and it's meant as a first step: see what AI-assisted computing looks like, then take it further on your own.

What's inside

Four ideas, one small tool

ARCade keeps the AI on a short leash: curated pipelines, tested jobs, and results you can inspect.

Plain-English chat

Type "Basecall and demultiplex mouse gut samples". No job scripts, no commands to memorise.

Pre-tested pipelines

Every job starts from a template we've tested on ARC. The AI tailors it to your request — your samples, your kit, your settings.

A local, open AI model

Google's Gemma model, tuned by us and running on a shared ARC server. No cloud AI service, no subscription, and your work stays on ARC.

Answers, not error logs

ARCade keeps an eye on your job, explains failures in plain English, and checks that the output looks the way it should.

How it works

From raw data to results in one conversation

For now, ARCade supports one workflow: Nanopore data preprocessing from the Griffin-Pipeline.

01

Put your data on ARC

Copy your raw sequencing files (e.g., .pod5) into a folder named for the run (for example, Mouse_gut_project), then point ARCade at it. It takes it from there.
02

Say what you want

ARCade follows your lead. Tell it what to run and it sets up the pipeline your way, checking with you on anything it still needs — such as your sequencing kit.
03

ARCade runs the pipeline for you

It sends your job to ARC with the right settings and keeps track of it, so you don't have to. Ask for status any time; spend the rest of your time on the science.
04

Results — or a clear explanation

When the job finishes, ARCade checks the output and sums up the results. Want something done differently? Did something go wrong? Just say so — it revises the script and runs it again.

Pipelines

Start small, grow carefully

ARCade runs a short, curated list of pipelines from the Bioinformatics Hub — each one tested end to end on ARC before we add it.

Long-read data preprocessing

From Nanopore raw signal (.pod5) to demultiplexed reads, with a quality-control report ready to open.

02 (Coming soon)

Taxonomic classification

Identify the microbes in each sample and estimate their relative abundance — picking up where preprocessing leaves off.

FAQ

Questions you might ask

What do I need to use it?

An ARC account, plus the University VPN when you're off campus.

Does my data leave the University?

No. The AI model runs on ARC, and your data stays there. ARCade doesn't call any cloud AI service.

In fact, ARCade never reads your data itself. It only works with file names, to run the pipeline, and with the output and log files, to check how the job went.

Is there sample data I can try?

Yes. We've shared raw Nanopore data (.pod5) from a ZymoBIOMICS mock microbial community — a commercial standard with a known composition, no human material — on Hugging Face.

Can it run my own pipeline?

Not yet — and that's the hard part of this experiment. Adding a pipeline to ARCade isn't just a matter of plugging in a script. The AI model has to learn what each step does, what inputs it expects, which settings matter, and what a good result looks like.

It also has to learn how the pipeline fails. A job that finishes without an error can still be wrong — the wrong kit, the wrong reference, too few reads — and ARCade needs to notice that, explain it in plain English, and know what to change. Every new pipeline means curating example conversations, training and testing the model, and running the whole thing end to end on ARC before we trust it with your data.

That's why ARCade starts with a short, carefully chosen list. The idea is simple: a research assistant you can talk to, running on your own cluster. Getting it right, one pipeline at a time, takes patience. If there's a pipeline you'd love to see next, tell us.

Found a bug?

Thank you for trying ARCade! This experiment is still a work in progress, so you may run into the odd bug — we appreciate your patience. Please let us know about anything you find; every report helps make ARCade better.

Get started

See ARCade live

Join us at the HUB workshop on November 6, 2026, where we'll walk through how to use ARCade together. Questions before then? Email us at Bioinformatics@ucalgary.ca.

ARCade, ARC aid via AI