The code behind the vaccine: how open source underpins personalized cancer treatment

Moderna and Merck recently reported Phase 3 clinical trial results showing that a personalized vaccine based on messenger RNA (mRNA), given alongside immunotherapy to patients with high-risk melanoma following surgery, significantly reduced the risk of the cancer returning. It was the first successful Phase 3 trial of an mRNA-based cancer treatment. 

It is not a preventive vaccine of the familiar kind, the sort given to healthy people to stop an infection before it starts. 

“This is a therapeutic vaccine,” explains Cédric Notredame, Professor of Computational Biology at MBZUAI and a member of the Scientific Advisory Board at Seqera—the company built around the Nextflow pipeline language, whose software was used to orchestrate the regulatory-grade computational pipelines behind these personalized treatments.  

“The way it works is that they take the patient’s tumor, sequence it, and use bioinformatics to figure out which specific bits of mutated proteins, what we call neoantigens, can be used to stimulate the patient’s own immune system to destroy only the cancer cells.” 

In principle, a patient gives a biopsy in the morning and, within a few days, a custom-synthesized molecular treatment is ready. The process is relatively cheap and simple, and relies on technology that already exists. What stands between the principle and the practice is computation. 

When code becomes a medical device 

Every personalized cancer vaccine requires its own genomic analysis. Rather than manufacturing a single molecule for millions of people, the work runs a separate, complex pipeline for each patient, ending in a distinct drug decision for one individual. 

“Because the algorithm determines exactly which RNA sequences go into that patient’s vaccine, the software itself is effectively making a drug decision,” says Notredame. “We have entered a new era of Software as a Medical Device.” 

At this scale, computational reproducibility is a technical preference. In biology, minor differences in operating systems or background dependencies can cause the same software running on the same data to yield different results in different hospitals. 

To solve that problem, Notredame and his former team at the Centre for Genomic Regulation in Barcelona developed Nextflow, an open-source scientific workflow language designed to run thousands of analyses reproducibly across any local or cloud environment. 

Notredame jokingly refers to Nextflow as the “HTML for pipelines.” Just as HyperText Markup Language (HTML) structures text so any browser can render it, Nextflow wraps disparate scientific software tools into scalable, reactive pipelines that can run on a laptop, a local cluster, or large cloud infrastructures without changing a single line of code. 

From academic spinoff to global standard 

From that same lab, Notredame’s students spun out Seqera, which now maintains Nextflow, provides enterprise support, cloud monitoring, and regulatory compliance around the Nextflow ecosystem. Moderna is a client of Seqera and uses this technology stack to orchestrate the genomic computation pipelines behind its personalized cancer vaccines. 

For some, a commercial enterprise whose main product contains zero patented intellectual property seems counterintuitive. Nextflow is fully open source; anyone can clone the code on GitHub. Yet Notredame argues that this open status is precisely why Nextflow was adopted where proprietary tools were not. 

“In the early 2000s, everyone doing genomics thought they were going to get rich, and everyone was protecting their code,” Notredame recalls. “But all the code that got protected immediately died. If you protect your code, people have to download it, sign papers, and they end up not using your stuff. If you want your work to be used, you make it open source.” 

Nextflow’s user base bears out Notredame’s assertion: Oxford Nanopore Technologies, whose sequencing technology supports the Emirati Genome Program; Genomics England, which is leading a national-scale whole-genome sequencing study of newborns; and the Darwin Tree of Life project, which aims to sequence the genomes of all eukaryotic species in Britain and Ireland. Nextflow is also routinely used across the pharmaceutical industry. 

Open source has also helped foster a substantial community around the software. nf-core, a global community that develops and maintains a collection of Nextflow pipelines, now brings together more than 15,000 people and continues to grow. 

The ethics of transparency in medicine 

Beyond the commercial advantage of fast, frictionless adoption, Notredame sees an ethical case for keeping scientific software open. 

“These algorithms are going to run on human lives,” Notredame stresses. “If a medical algorithm is a black box, nobody can verify or argue against it. That’s not science.” 

As Editor-in-Chief of the journal NAR Genomics and Bioinformatics, Notredame enforces this standard, requiring all published pipelines to be open source and unrestricted. 

As the world moves toward personalized medicine, the open-source software holding this medical infrastructure together must be actively maintained and protected from silent decay. This is why Notredame and other scientific leaders are advocating the creation of an Open Science Software Foundation, loosely modeled on the Linux Foundation, to ensure critical bioinformatics tools remain transparent, secure, and robust. 

“From a small lab tool to powering the frontier of cancer medicine,” says Notredame. “This is just the beginning of what computational biology can achieve.”  

Bringing Nextflow to deep phenotyping 

Notredame is now bringing Nextflow to MBZUAI, where the same infrastructure is intended to support the Human Phenotype Project, the large-scale deep-phenotyping effort led by Eran Segal, Dean of the Biological and Life Sciences Division, and Professor of Computational Biology at MBZUAI. 

The computational problem is structurally the same as the vaccine one: many individuals, each generating a dense stream of measurements across genomics, imaging, metabolism, and the microbiome, each requiring analyses that have to run identically today and five years from now. What changes is the question. Instead of asking what a single tumor has mutated into, deep phenotyping asks what a healthy body does over time, and how its signals shift before disease appears. 

“The pipelines are the experiment,” says Notredame. “If you cannot rerun them, you cannot compare a participant measured in 2026 with the same participant measured in 2031, and a longitudinal cohort stops being longitudinal.”