Hackathons, Bioinformatics, and being in Australia

The first hackathon I ever signed up for was in 2019. I thought I’d written about that particular experience, but it seems I only wrote about previous experiences that were successful-ish, which is weird. Anywho, the short summary of that first ever is that I lasted only a few hours before I was, quite literally, kicked out. Harsh. But also funny, and it left me curious. What is the point of hackathons anyway, and why was I removed instead of being left to drift as the “typical inactive bystander” you find in most online communities?

Because I’m stubborn, I have since participated in five or more others. Three of them happened during my PhD, and those are the ones I want to talk about. Looking back, what they actually taught me had little to do with the science. They taught me collaboration. And I finally understand why I was kicked out of that first one.

The setup

Hackathon map of participants. Source: one of the hackathon pages

All 3 of these hackathons were organised by people or groups based in the United States. I was often the only team member in Australia. Anyone who has tried to collaborate across that particular timezone gap knows that “active participation” in the conventional sense, as in being on calls, jumping in on Slack live, is mostly off the table. I knew that going in, so all I had to do was figure out how to participate.

What I did, across all three, was take on the work that did not require real-time presence. I volunteered to be the team’s writer and quality control person. I could not write convincingly about something I had not understood and verified, so writing forced me to read every line of code, run every pipeline, and confirm that the outputs matched what the team claimed they did. In effect, the role made me the team’s first external user. I read the repo the way someone arriving in six months would, looking for what was missing, what was unclear, what would break the moment someone tried to reproduce it.

I still joined meetings at 2 a.m. when I was awake. Just not all of them, and of course, I communicated this beforehand.

Two of the hackathons ran across three days. The third was a couple of weeks long.

Hackathon 1: Machine learning and AI approaches to multimodal problems in computational biology

Official page: https://guides.library.cmu.edu/hackathon/start, GitHub: https://github.com/collaborativebioinformatics/Cis_and_trans_effects_on_variant_penetrance

Five of us worked on a project titled cis and trans effects of haplotypes on rare variants. The direction was led by our team lead and went heavily into population and ancestry methods — not my area, and not where my interest naturally sits. Honestly, I had to work to stay engaged with the framing. The team adapted StructLMM, a linear mixed model originally built for gene–environment interactions, to detect gene–gene interactions by using local ancestry principal components as a proxy for structured environmental effects. I cannot claim I followed every modelling decision. I can claim that the code worked because I tested it.

We won Best Documentation. I take real pride in that.

The README I structured told you exactly where to download the data, how to set up the environment, what to expect at each step, and where to look when something failed. It is what makes a piece of research code survive the moment its authors move on, and at that hackathon, it was the contribution I could make that mattered most.

I also stayed on after the hackathon ended to contribute to the manuscript. That part I want to flag specifically, because it’s very hard, if not impossible, to fully complete a project in 3 days. The teams whose work persists are the ones where at least one person keeps showing up after the deadline.

Hackathon 2: SV Hackathon 2025

Resource: https://fritzsedlazeck.github.io/blog/2025/hackathon-2025/ ; GitHub: https://github.com/collaborativebioinformatics/MethSmoothEval

I think this should’ve been called SV and long read hackathon. The project I joined had nothing to do with SVs and everything to do with long-read methylation, which is the closest a hackathon has ever brought me to my own PhD work.

The question we asked was whether smoothing — a technique inherited from short-read methylation analysis for technical reasons — is appropriate when applied to long-read CpG signals, particularly when you care about differentially methylated regions and individual CpG sites rather than broad averages. We worked with publicly available cell line and tissue data, and much of what we found pointed in the same direction: smoothing should be used cautiously in long-read contexts, especially when the analysis depends on single-CpG resolution.

I played a similar role here. Documentation, QC, testing the codebase as someone arriving fresh would, and contributing to the manuscript that is still being prepared. I also contributed to the analysis and result generation to an extent.

What this hackathon actually did for me was retroactive. When I was writing the code to calculate methylation differences for my own project, smoothing was everywhere in the literature and in the tools whose code I was studying. My instinct kept rejecting it. I did not want the smoothed version; I wanted the real signal, but I could not fully articulate why beyond that. This hackathon gave me the justification I had been operating without.

Hackathon 3: Research to the people Lung fibrosis case

Resource: https://researchtothepeople.notion.site/lung-fibrosis

I did not sign up for this one through the usual channels. A member of Biospherea, a community we both belong to, posted that they needed someone familiar with long-read data. I thought, why not. The data turned out to be PacBio DNA and RNA, a platform I had no hands-on experience with, and the rest of the team focused on a few other modalities. The format was also different from the others. It was a real patient. The organisation was, in plain terms, asking hackers to dig through the data and look for anything that might help reverse the condition.

The typical analyses, like variant calling and isoform calling, had already been done. My contribution was to annotate the called variants, manually inspect the reads supporting each in the gene of interest, and try to bridge that to the RNA data, knowing the DNA and RNA samples were from different sources (blood and hair) and would not align cleanly. I was looking for any consistency between variant presence and the transcriptomic signal.

I did not find much that could directly help the patient. I did find what I’d call “real-life” evidence of false positives, which is a widely documented issue in clinical genomics. Several variants that had passed the standard filters were not visible in the underlying reads when I checked.

The team’s broader findings went down an immunology angle that is well outside my expertise, so I won’t pretend to summarise it. But it was the closest I have ever been to a clinical question, and that mattered.

Hackathon equals high-energy collaboration

Across 3 hackathons, I never wrote the headline code. I was never the team lead. I was geographically alone and outside the room where most of the decisions were made. And it did not matter. Documentation and QC are the roles/tasks most likely to be skipped under deadline pressure, and the roles most responsible for whether the work survives the deadline. Thankfully, my situation made me the right person to play those roles.

The other lesson is about persistence. In all 3 cases, the project did not end when the hackathon did. Because, unless you’re fine with the foundation or driving principles behind the work disappearing, you need to hang around long enough to create informative manuscripts.

Finally, being in the right communities is part of how any of this happens at all. None of these 3 hackathons came to me through formal channels. The first 2 I found on a living biohackathon page, and I don’t even remember who shared that page with me. One came through a community I had joined out of curiosity. The opportunities exist, and being available to them is its own skill.

In that first hackathon in 2019, I was waiting to be told what to do. The point of a hackathon is that no one is going to.