Bio2Byte information

The Bio2Byte group is led by Prof. Dr. Wim Vranken, with many people contributing to the research.

If you are interested in our work, keep an eye on the publications in which the lab has been involved.

For any further questions, feedback or suggestions, please send an email to: bio2byte@vub.be.

Stay updated with our latest news, projects, and collaborations by following us on our social media and platforms:

• LinkedIn
• X (Twitter)
• Zenodo
• GitHub
• Bitbucket

For joining us, have a look at what possible opportunities we have.

The Bio2Byte group is primarily situated at the Interuniversity Institute of Bioinformatics in Brussels, a collaborative interfaculty institute between the Vrije Universiteit Brussel and the Université Libre de Bruxelles. It is located at the the ULB side of the Pleinlaan/La Plaine campus on the 6^th floor of the C building.

At the VUB, the group is linked to Structural Biology Brussels at the Bioengineering sciences department, as well as the departments of Computer Science and Chemistry at the Faculty of Sciences, plus to the Biomedical sciences at the Faculty of Medicine and Pharmacy.

Our research is funded from various sources, listed below:

• PROHITS (Jan. 1, 2024-Dec. 31, 2027) from EC
  

  PROHITS is a MSCA doctoral network (DN) action coordinated by us. The overall aim of PROHITS is to integrate experimental and computational approaches that can resolve the proteome of prokaryotes at single cell level, and at a range of temperatures, in order to improve our understanding of thermophile biology and to guide the optimized design of cell factories. For more information, please go to https://prohits.eu.

• FWO-IRI ELIXIR (Jan. 1, 2023-Dec. 31, 2026) from FWO
  

  Infrastructure grant from the Flemish government for ELIXIR-Belgium, where we receive funding to develop services for the 2023-2026 period.

• O&O project with Confo therapeutics (Jan. 1, 2023-Dec. 31, 2024) from VLAIO
  

  We receive funding from this O&O project to connect our tools to inside approaches at Confo Therapeutics to help with the in silico screening of new target proteins.

• FWO-SB (Nov. 1, 2022-Oct. 31, 2026) from FWO
  

  FWO-SB grant to Charlotte Crauwels to work on the design of hybrid GPCRs

• Synthetic biology SRP (Oct. 1, 2022-Sept. 30, 2027) from VUB
  

  Strategic Research Project (SRP) from the VUB with as topic 'Synthetic biology for the development of microbial cell factories'. Other partners are Eveline Peeters and Sophie de Buyl, both VUB.

• PhD grant (May 1, 2022-April 30, 2026) from FNRS
  

  Grant for Sophie-Luise Heidig from the FNRS in a joint PhD with the ULB (Francois Flot).

• Hosmart-AI (April 1, 2021-March 31, 2025) from EC
  

  Hosmart-AI is a European project to develop a marketplace for AI solutions in the health sector. We contribute one of the 'pilots' in this grant, based on TumorScope.

• ProteinContour (Jan. 1, 2021-Dec. 12, 2024) from FWO
  

  In this project (FWO identifier G028821N), we combine mass spectrometry analysis expertise from UGent (CompOmics group, Lennart Martens) and UCL (Laurent Gatto) with our predictions to map the relationships between post-translational modifications, sub cellular location, interactions and biophysical features of proteins in human cells.

• Implementation studies (Jan. 1, 2021-May 31, 2023) from ELIXIR
  

  We participated in several ELIXIR implementation studies (IDP, IDPTools), with focus on creating connections between the IDP and other ELIXIR communities (3DBioinfo, proteomics).

• TumorScope (Oct. 1, 2019-Sept. 30, 2024) from VUB
  

  In this interdisciplinary research project (IRP) from the VUB, we leverage the unique expertise and data present at the VUB and the UZ Brussel to create a general framework to store and analyze raw medical data, both at the image and molecular level, in relation to solid tumors, their clinical behavior and response to therapies. An integrated research platform, which conforms to GDPR and patient legislation, will enable AI-driven extraction of new information on breast and brain tumors, as well as sarcomas. Combined with expert insights from clinicians, we will work towards assisting and improving the decision-making process in especially multi-disciplinary oncology consults (MOC).

• RNAct (Jan. 1, 2019-Dec. 31, 2023) from EC
  

  The H2020-MSCA-ITN-2018 813239 RNAct project focusses on 'Enabling proteins with RNA recognition motifs for synthetic biology and bio-analytics'.

• FWOESFRI2 (Nov. 6, 2017-Dec. 31, 2019) from ELIXIR-Belgium
  

  Bringing European standards to Brussels : Preparing the human health related (IB)2 services and databases for ELIXIR.

• G.0328.16N (Jan. 1, 2016-Dec. 31, 2019) from FWO
  

  FoldMod - The role of local amino acid interactions in the folding of proteins, the stability of their fold and the location of post-translational modifications.

• ICITY-RDI.BRU (Jan. 1, 2014-Dec. 31, 2023) from ICITY
  

  A software framework for identifying molecular causes of human diseases (Mutaframe)

• Ph.D. grant Daniele Raimondi (Jan. 1, 2013-Dec. 31, 2017) from IWT
  

  Advancing the prediction and understanding of coding variant effects

• BB2B 2010-1-12 (Dec. 1, 2010-Nov. 30, 2015) from Innoviris
  

  Brains Back to Brussels project - Combining experiment and informatics to study biopolymer dynamics and interactions.

A summary of platform changes, most recent first.

v20260817.1 (deployed on 2026-08-17)

The Biophysical Proteome Atlas now covers several species instead of only human. Alongside human it holds the reviewed (Swiss-Prot) entries of the baker's yeast, zebrafish, fruit fly and thale cress reference proteomes, each carrying the same five sequence-based predictions, so results are comparable between proteins and between species. Each proteome now has its own pages and entry list, predictions for the new proteomes are computed the first time somebody opens an entry, and the atlas gained a quick search, a REST API, a status page and a redesign throughout.

• Existing links still work. Entry addresses did not change — /proteome/protein/<accession> resolves as before, whichever organism it belongs to.
• The unreviewed (TrEMBL) human entries have been retired, since manual curation is what makes entries comparable across species. Nothing curated was removed, and a link to a retired entry now explains what happened and offers to run the same predictors on that sequence for you.

v20260813.1 (deployed on 2026-08-13)

• ACPYPE entry pages are considerably faster, and no longer show “There was a problem reading the selected entry” when the site is busy.
• Protein pages in the Biophysical Proteome Atlas are smaller and load in a fraction of the time.
• Sorting the protein list by number of residues is no longer slow. The residue count is now stored with each entry instead of being measured across the whole database on every request.

v20260810.1 (deployed on 2026-08-10)

• The Online Predictors REST API can now be used from your own scripts. Submitting a prediction from outside a web browser previously failed with a “CSRF verification failed” error.
• Submissions accept a JSON request body, as the documentation had always described.
• Checking on a prediction now tells you what happened from the response code alone: still running, finished, or failed.
• A failed prediction now reports why it failed through the API.
• Added a fair-use limit of 100 submitted jobs per hour per token, counted over a rolling hour and shared across all three submission endpoints. Rejected submissions do not count towards it, and every accepted submission reports how much of your allowance is left. If you need a higher allowance for a research project, please get in touch and we can lift the limit for your token.
• Rewrote the Online Predictors API documentation. It now includes ready-to-run curl commands for every endpoint, and a complete worked example in both Python and shell.
• Fixed jobs that appeared to run forever. If a prediction's worker was stopped — by the time limit, by running out of memory, or by a restart — the job stayed marked as “running” indefinitely, with no error and no way to tell it apart from one still being worked on. A running prediction now reports that it is alive every few seconds, and one that stops reporting is marked as failed with the reason.
• Abandoned jobs are now cleaned up on a schedule, not only when someone opens the page they are on. A job whose worker stopped is marked as failed within a few minutes either way, so the queue figures on the monitoring page stay honest — and that page now also reports whether the scheduled clean-up is itself running.
• A prediction that runs out of time is no longer retried. It used to be attempted three times, each one allowed the full time limit, so a file too large for the platform took around 45 minutes to report a failure that was clear after the first attempt. The failure now says so directly and suggests submitting fewer sequences or fewer predictors.
• The time limit is now actually enforced. A prediction that ran over its limit used to keep going: the interruption was absorbed by the prediction library, which skipped the sequence it was working on and moved to the next one, so the job ran on and eventually failed with an internal message that had nothing to do with running out of time. Predictions are now stopped when their time is up, on both the current and the legacy tools, and the failure reported is the timeout itself — so a job too large for the platform says so, instead of reporting a confusing error much later.
• Legacy tool submissions (MSA, PSP and ShiftCrypt) no longer stay at “processing” for ever when their worker is stopped. As on the current platform, a running job reports that it is alive, and one that stops reporting is marked as failed with the reason. Jobs still waiting for a free worker are left alone: a queue is not a failure.
• Your My predictions page now shows how long each job waited in the queue, how long it has been running, and how much time it has left, updating while it runs. Previously a single “elapsed time” appeared only once a job had finished, and it silently counted the queue wait as part of the prediction. The same figures are available from the API.
• The three prediction types now behave identically. Whichever predictor you use, and whether you use the website or the API, the same problem is reported in the same words, a submission answers with the same fields, and the same information appears in the same place. Previously a missing token produced three different messages, only ShiftCrypt could report what a job was doing, and only single-sequence jobs recorded the task that ran them.
• Each job now keeps a processing log, reachable from the “Processing log” button on your predictions and through the API. It lists what the job did and when — queued, started, each phase of the prediction, finished or failed — which makes it possible to see where a long job actually is.
• Refreshed the look of the site: one consistent colour scheme throughout (parts of the interface were still using the default framework blue rather than the Bio2Byte green), clearer tables that keep their headers visible while you scroll sideways, visible keyboard focus, and reduced motion for anyone whose system asks for it. Stylesheets are now versioned, so a visual fix reaches returning visitors instead of being hidden by their browser cache.
• Your predictions now show a Stage column saying what a job is currently doing — “Running predictions”, “Saving results”. All three predictors were already producing these descriptions internally and discarding them.
• Single-sequence and MSA jobs now record their state changes, so the “View History” panel on a result page is no longer empty for them.
• Corrected the time limits quoted on the predictions page, which said 20 minutes for a 15-minute limit.
• Added guidance on the submission pages about how much one job can do, so a very large file can be split up before it is submitted rather than after it fails. Submissions are still not limited by size.
• Fixed ShiftCrypt submissions that carried no input data leaving an empty, permanently queued job behind instead of being rejected outright.
• An over-long token is now rejected immediately with a clear message, rather than being accepted and then failing confusingly later.

v20260806.1 (deployed on 2026-08-06)

• New Online Predictors platform (/online_predictors/) alongside the existing legacy tools, running on the same Celery/worker infrastructure, with job naming, a system status page, and workers tuned for heavier real-world load.
• Consistent failure reporting across all prediction APIs, so a failed job reports the actual reason instead of a fabricated result or a generic timeout message.
• Redesigned ACPYPE (topology builder) submission pages, with clearer error, not-found and internal-error pages.
• Fixed the multiple sequence alignment (MSA) tutorial plot on the legacy platform, where the median, prediction and legend lines could silently fail to render.
• Fixed downloading result files (JSON, CSV, metadata, input, distributions) on the Online Predictors platform.
• The "Submit & start a new one" button on the Online Predictors platform now keeps the selected analysis type (Single Sequence / MSA / Chemical Shifts) instead of resetting it.
• Corrected outdated citations, including the DisoMine reference (now pointing to its published Journal of Molecular Biology article) and the predictor references on the Online Predictors "About" page.
• Added an end-to-end integration test suite exercising both prediction platforms against the real queue/worker stack, to catch regressions before deployment.