Seqera, the company behind the open-source Nextflow software, launched Nextflow Frontier on 13 October 2026 [1]. Seqera describes it as “a recursive closed-loop discovery system for the unsolved problems in life sciences”. In plain terms, it is an AI system that comes up with ideas, tests them as computer experiments, checks the scores and tries again.
What Nextflow is
Nextflow is free software that scientists use to write down a chain of computer steps, such as cleaning data, running an analysis and training a model, so the whole chain can be rerun the same way on any computer or cloud. These chains are called pipelines. It is widely used in biology to process large amounts of data, such as DNA sequences. Seqera says tens of thousands of scientists use Nextflow every day and that more than four million workflows run on it every month [1].
How Frontier works
Seqera says a team gives Frontier a problem and “a clear way to measure success” [1]. The system then comes up with possible approaches, turns each one into a Nextflow pipeline and runs them, often thousands at once. It scores the results, learns from them and runs again. Seqera says it can run on its own for many days while scientists can still guide and check it, and that it works with whichever AI model gives the best results at each step.
Seqera lists several parts: a control panel where scientists set the problem and follow progress; a “research harness” with AI agents for reading scientific papers, analysing data and writing software; “Agentic Nextflow”, which Seqera calls a proprietary version of Nextflow built for very large, long-running jobs; a planning model that decides what to try next; connectors to data and lab systems; and Seqera’s existing Nextflow Platform [1].
What Seqera says it achieved
All of these results come from Seqera and have not been independently verified.
- BioML-bench. This is a set of 24 tasks built on data from published lab experiments, covering drug discovery, medical imaging, single-cell biology and protein engineering. Seqera says earlier AI systems did better than about 75% of the human leaderboard entries on average, while Frontier has done better than 92% [1]. Seqera says Frontier has hit top results, or the statistical equivalent, on 17 tasks, and 7 are still running.
- EGFR. EGFR is a protein that can drive cancer. Seqera says it gave Frontier 500 molecules already tested against EGFR in the lab and held back another 144 [1]. Over 80 hours, Seqera says, the system tried 92 ideas and ran 800 scored pipelines, and kept pace with the best public leaderboard entry using a chemistry model small enough to run on a laptop. Its top three picks from the 144 held-back molecules all share a piece of chemistry with lapatinib, an existing EGFR drug. These molecules came from the test set. They are not new drugs or new drug candidates, and Seqera says so itself.
- Other tasks. Seqera says Frontier beat a rebuilt version of the winning approach at predicting how fast lung function declines in pulmonary fibrosis, though it notes this was on the same patients it was tuned on [1]. It also says Frontier named kidney cells more accurately than the published leader, matched protein stability results with a model 80 times smaller, and pulled in rat liver data from the same public study to help predict how fast the human liver breaks down a drug.
One customer is quoted. André Faure, co-founder and chief technology officer of ALLOX, says that on his company’s own held-out data, the model Frontier built “outperformed every published method we tested” at predicting how mutations change protein binding [1].
Who it is for
Seqera says Frontier “is not sold as software; it’s a partnership” [1]. Each project starts with a conversation, and Seqera supplies the system plus its own scientists. Seqera says partners own the new models and methods it delivers. It is working with “a select set of advanced life sciences teams” for now. No pricing was given.
Seqera also says 15 of the top 20 pharmaceutical companies use its platform, that Anthropic has built Nextflow skills into Claude for Life Sciences, and that OpenAI built its Rosalind Workbench on nf-core, a community collection of Nextflow pipelines [1]. These are Seqera’s statements.
What this means
- It is a service, not a download. Companies have to partner with Seqera, and the price is not public.
- The results are Seqera’s own. Nobody outside the company has checked the benchmark numbers yet.
- It finds methods, not medicines. The EGFR test ranked known molecules. It did not create new drugs.
Sources
- Seqera (Evan Floden), “Introducing Nextflow Frontier”, 13 October 2026. https://seqera.io/blog/introducing-nextflow-frontier/
