Themes & takeaways — EuroBioC 2025¶
Cross-day synthesis of the recurring threads, surprises, and named entities across Day 1, Day 2, and Day 3. EuroBioC 2025 was a 3-day, ~5-keynote / ~25-short-talk / 5-workshop event in Barcelona — the European leg of the Bioconductor calendar, tighter and more methodological than the joint GBCC counterpart. The themes below are what the program converges on once you read the schedule as one document rather than three.
Source caveat: Plenary recordings are public on the Bioconductor YouTube playlist; short-talk slides typically land on Zenodo within a few weeks but were largely TBD at recap time. Claims below are anchored to talk titles, package READMEs, and the official community recap — recordings will tighten the specifics.
The five threads¶
1. Spatial-omics convergence¶
The single loudest thread of the meeting. The European Bioconductor scene is consolidating fast around spatial methods, and Day 2 was where it surfaced as a coordinated stack:
- Helena Crowell's Day 1 keynote anchored the case for whole-transcriptome imaging in colorectal cancer.
- SpatialData (Marconato) provides a Python-primary data structure with an in-development R interop layer — the meeting's clearest acknowledgment that spatial data formats are now cross-language by default.
- DESpace (Cai) layers spatial differential expression on top of edgeR's NB machinery — i.e., a classical Bioconductor workhorse extended with spatial context.
- spatialFDA (Emons) pushes the same problem into the multi-sample regime where clinical use lives.
- OSTA (Day 2 workshop) — the Orchestrating Spatial Transcriptomics Analysis book, taught in person, locks the curriculum.
Five contributions, four different speakers, one consistent direction. This is what Bioconductor convergence actually looks like in real time.
2. Cross-language interop as a first-class concern¶
Two talks foregrounded language bridges in their titles, not as footnotes:
- SpectriPy (Rainer) — explicit R↔Python mass-spec interop, named for the bridge.
- SpatialData — Python-primary, R interop in development, openly acknowledged.
The implicit version of the same thread runs through the Workshop Platform (containerized environments that don't care which language), Metabonaut (a tutorial framework that has to interoperate with SpectriPy), and the broader pattern of Python-primary frameworks (anndata, scanpy, MuData) that the R-side is increasingly mirroring rather than competing with. The Bioconductor stance has shifted from "we are the R ecosystem" to "we are the R-anchored interoperable ecosystem."
3. Latent-variable / dimensionality-reduction methods proliferate¶
Statistically, this was the year's centerpiece — and it was visible across both days:
- Susan Holmes' keynote — Latent variables as the best medicine for heterogeneity — argued the methodological frame head-on.
- omicsGMF (Clement) — generalized matrix factorization usable across modalities.
- consICA (Chepeleva) — independent-component analysis with a consensus / stability layer.
Three contributions, three different angles (latent factor models, GMF, ICA), one common claim: when the biology is heterogeneous, the right primitive is a latent representation. Pair this with the spatial track and the picture sharpens — spatial heterogeneity is heterogeneity, and the same tools apply.
4. Proteomics renaissance from the Clement lineage¶
A quieter but striking thread: multiple proteomics-flavored tools traceable to Lieven Clement's group, with complementary roles:
- omicsGMF (Clement himself, Day 1) — dimensionality reduction.
- msqrob2 (Vanderaa, Day 2 workshop) — robust statistical proteomics, the workhorse downstream of search engines.
- Lucas Beerland's distributional-inference single-cell proteomics talk (skipped from the tools index but in the program) — the experimental frontier of the same lineage.
Bioconductor proteomics is having a moment, and one Belgian group is producing much of the gravitational mass. This complements the metabolomics track (notame, Metabonaut, SpectriPy) — the small-molecule side is healthy too.
5. Workshop Platform + BiocBook = the meta-infrastructure¶
The most strategic thread runs underneath the talks:
- Workshop Platform (Mahmoud, Day 2 workshop) — the containerized environment that runs all the workshops in the room, deployable as a per-event service.
- Jacques Serizay's Day 3 keynote — community-driven flexible software, BiocBook, the ecosystem-as-coordination-mechanism argument.
- OSTA as a book taught as a workshop on a platform — the three layers compose.
The closing argument of the meeting is that the value Bioconductor delivers is not packages, it's the production line that turns research into a reproducible, shareable, runnable artifact. Workshop Platform is to a workshop what BiocBook is to a vignette is to a peer-reviewed package — three rungs on the same ladder.
Surprises worth flagging¶
| Surprise | Day | Why it stood out |
|---|---|---|
| Five distinct spatial-omics tools across two days | 1 + 2 | The European scene is now coordinating on spatial, not competing |
| SpatialData as a Python-primary "Bioconductor" talk | 2 | Explicit acknowledgment that the data-structure standard isn't R-only |
| SpectriPy named for its language bridge | 1 | Cross-language interop is a title-worthy contribution now |
| Holmes keynote framing latent variables as "medicine" | 1 | Methodological argument, not a results talk — the rhetorical centerpiece |
| edgeR still getting a methods-talk slot 20 years in | 1 | The classical workhorses are not done; Lizhong Chen's update went alongside the new spatial-DE tools |
| Workshop Platform as a workshop about workshops | 2 | The infrastructure-of-infrastructure show, quietly the most leveraged talk in the room |
| No agentic-AI / LLM track at all | all | In contrast to Nextflow Summit 2026; see "notably absent" below |
Tools, keynotes & entities — quick index¶
| Item | Day 1 | Day 2 | Day 3 |
|---|---|---|---|
| Vince Carey — 25 years of Bioconductor | ✓ | ||
| Helena Crowell — CRC imaging | ✓ | ||
| Susan Holmes — latent variables | ✓ | ||
| James Sharpe — C3PO | ✓ | ||
| Noelia Ferruz — protein LMs | ✓ | ||
| Jacques Serizay — community software | ✓ | ||
| notame | ✓ | ||
| Metabonaut | ✓ | ||
| SpectriPy | ✓ | ||
| omicsGMF | ✓ | ||
| miaTime | ✓ | ||
| RCX | ✓ | ||
| MIRit | ✓ | ||
| DeeDeeExperiment | ✓ | ||
| edgeR | ✓ | ||
| MOSClip | ✓ | ||
| Rega | ✓ | ||
| footprintR | ✓ | ||
| decemedip | ✓ | ||
| SpatialData | ✓ | ||
| DESpace | ✓ | ||
| spatialFDA | ✓ | ||
| consICA | ✓ | ||
| barbieQ | ✓ | ||
| OSTA | ✓ | ||
| mia | ✓ | ||
| msqrob2 | ✓ | ||
| Workshop Platform | ✓ | ||
| iSEE | ✓ |
Speakers / institutions named in the program: Vince Carey (Channing/Harvard), Helena Crowell (CNAG), Susan Holmes (Stanford), James Sharpe (EMBL Barcelona), Noelia Ferruz (CSIC), Jacques Serizay (Institut Pasteur), Lieven Clement (Ghent), Christophe Vanderaa (Ghent), Charlotte Soneson (FMI/SIB), Federico Marini (Mainz), Luca Marconato (EMBL), Tuomas Borman (Turku), Johannes Rainer (Eurac), Alexandru Mahmoud (Bioconductor core), and the short-talk roster captured in the tools index.
Domains represented: spatial omics, single-cell, microbiome, mass spec / metabolomics, proteomics, RNA-seq, miRNA-mRNA, multi-omics survival, methylation, networks, clonal tracking, EGA submission, visualization, infrastructure / data classes, books / training.
What's notably absent¶
- No agentic-AI / LLM-driven-pipeline track. The Nextflow Summit 2026 vault is dominated by agentic AI threads — Co-Scientist, MCP-per-pipeline, LLM-driven QC rewrites. EuroBioC 2025 has one AI keynote (Ferruz on protein design) and effectively zero short talks on agentic tooling for Bioconductor itself. Striking, given the topic's gravity elsewhere; an opening for BioC2026.
- Light on benchmarks-of-benchmarks. Most short talks present new tools; few revisit the comparative-benchmark genre that the AACR / ISMB methods circuits have been pushing. The Bioconductor review process front-loads this work into vignettes, but a session-format methods-comparison track is missing.
Where the corpus connects¶
- AACR 2026 (../aacr-2026/) — the bioinfo-tools and single-cell/spatial-omics topics share a substantial subset of the spatial stack here (SpatialData, DESpace, OSTA). When AACR speakers cite "we used SpatialExperiment / SpatialData" — this is the meeting where the underlying tooling got versioned.
- Nextflow Summit 2026 (../nextflow-2026/) — opposite framing: Nextflow is pipelines / orchestration / agents, EuroBioC is methods / packages / statistics. The two are complementary halves of the bioinformatics stack — Nextflow ships the spatial-preprocessing pipeline, Bioconductor ships the spatial-DE method that consumes its output. The contrast on agentic AI (heavy at Nextflow, near-absent here) is the most legible cross-vault signal.
- GBCC 2025 (queued sister vault) — the Galaxy + Bioconductor joint event. The OSTA / mia / iSEE workshops in particular will likely show up there in re-taught form; cross-linking once GBCC is populated.
- JPM 2026 (../jpm-2026/) — distant from this vault on content but worth noting that the Crowell-keynote disease focus (CRC, whole-transcriptome imaging) sits in a space that AI-pharma announcements are starting to circle (multi-modal foundation models for oncology — see Modella AI). Methods first, then deals.
Caveat¶
EuroBioC 2025 has a public schedule, public plenary recordings, and a public community recap; short-talk slides land on Zenodo on a multi-week lag. This synthesis is built from titles + package documentation + the recap blog, with a small number of inferences explicitly flagged ("the talk title suggests…"). Recordings should be the source of truth where the precise methodological claim matters.