why don’t biologists use openreview?
{needs actual context in the beginning}
Bench biologists generally do not use OpenReview for ordinary conferences. OpenReview is common in machine learning and adjacent computational fields, and it has leaked into computational biology, bioinformatics, biomedical informatics, and AI-for-health venues. For example, ACM BCB 2025 and 2026 have OpenReview pages, and IEEE BHI 2025 explicitly advertised OpenReview for regular conference papers. (OpenReview) But that is not the modal ASCB, GRC, CSHL, Keystone, FASEB, Gordon-style bench-biology meeting.
The big historical split is: computer science made conferences into archival publications; biology made conferences into social and scientific coordination events around work that will later be published in journals. ML inherited the CS model, then intensified it. In CS, conferences are often treated as the primary publication venue, with peer-reviewed proceedings, acceptance rates, citations, and hiring value. Springer Nature describes proceedings as central in CS and says top conference proceedings can carry more prestige than journals; bibliometric work also finds that CS values conferences more highly than other fields. (Springer Nature)
That one difference explains a lot of the surface weirdness. In ML, submitting to NeurIPS, ICML, or ICLR means submitting a full paper for a publish-or-reject decision. The conference is doing journal-like filtration. OpenReview then becomes a natural infrastructure layer: public manuscript pages, anonymous official reviews, rebuttals, meta-reviews, discussion threads, and post-decision public records. ICLR 2026, for instance, says official reviews are anonymous and publicly visible in OpenReview and that logged-in users can post public comments during discussion; NeurIPS 2025 says accepted papers and opted-in rejected papers are made public with reviews, meta-reviews, author responses, and reviewer responses. (ICLR)
In bench biology, a conference abstract usually does not play that role. At Cell Bio 2026, ASCB explicitly says abstracts are not peer-reviewed, although they are published online and as a citable supplement. (ASCB) CSHL says the organizing committee reads abstracts and decides oral versus poster placement after the deadline. (Cold Spring Harbor Laboratory Meetings) FASEB-style conferences similarly use abstracts to select short talks while still expecting poster presentation. (FASEB) That is program curation, not archival peer review.
Gordon Research Conferences are the sharpest contrast. GRC policy treats presented material as private communication, prohibits recording and quotation without permission, and says scientific publications should not be prepared as emanating from the conferences. (Gordon Research Conferences) That is almost the anti-OpenReview cultural template: unpublished work is deliberately shown in a semi-private trust environment.
Mechanistically, the difference comes from several field-level constraints.
First, ML papers are unusually reviewable as PDFs. A reviewer can inspect the method, theorem, benchmark, ablation, code link, dataset, and comparison table. The work may still be wrong, gamed, or under-replicated, but the object being judged is close to the final scholarly object. In bench biology, the conference object is often a 250 to 3,000 character abstract plus a poster. The real epistemic weight sits in raw images, gating strategies, replicate structure, animal cohorts, reagent validation, sequencing QC, perturbation design, antibody specificity, cell-line history, and the ugly negative controls that do not fit into an abstract.
Second, ML has faster experimental half-lives. A training run, benchmark suite, or model variant can be packaged on a conference cycle. Wet-lab biology is constrained by cloning, cell culture, animal aging, imaging schedules, failed antibodies, batch effects, vivarium delays, IRB or IACUC constraints, and reviewer two asking for the one experiment that takes four months. Biology therefore routes final adjudication through journals, where revision is part of the expected process.
Third, bench conferences often trade on controlled disclosure. People show unpublished data because they want feedback, collaborations, priority signaling, recruitment, and gossip-filtered field intelligence. Full public review before the paper is ready would make many PIs worse off: it can burn novelty, create patent risk, expose half-built stories, or produce citable criticism of a result still being debugged. ML solved part of that by normalizing arXiv and public priority timestamps; bench bio has adopted preprints unevenly and still has stronger norms around incomplete unpublished data.
Fourth, ML has a brutal centralized prestige economy. A NeurIPS, ICML, or ICLR acceptance is a career token, a hiring signal, an investor or industry visibility signal, and a benchmark race marker. That produces huge submission volume, standardized deadlines, massive reviewer pools, rebuttal rituals, and increasingly bureaucratic meta-review. OpenReview fits that world because the conference is a publication market. Bench bio prestige is more distributed across journals, lab pedigree, invited talks, society awards, grants, and who trusts your data.
The interesting edge case is computational biology. ISMB is much closer to the CS model: its proceedings are full citable articles in Bioinformatics, and the 2024 ISMB proceedings report 58 accepted papers from 318 full-length submissions, with 4.38 reviews per paper on average. (OUP Academic) So comp bio, biomedical informatics, and AI-for-health can look ML-like, while ASCB or GRC look bench-like.
Life sciences do have open-review experiments, but they usually attach to preprints and journals rather than conferences. Review Commons peer-reviews life-science preprints before journal submission; eLife publishes reviewed preprints with public reviews and an assessment; PREreview hosts open preprint reviews. (Review Commons) That tells you the cultural pressure for transparency exists, but it is being mapped onto the journal/preprint axis rather than the conference axis.
So the clean answer is: bench biologists mostly do not use OpenReview because bench-biology conferences are not usually the field’s main publication filter. ML conferences look different because they are simultaneously journal, priority registry, leaderboard checkpoint, job-market signal, and annual coordination event. Bench biology conferences are more often a mixture of unpublished-data salon, society meeting, poster market, PI-status theatre, and pre-journal feedback loop.