I have created Bioicons - a library of 1632+ free, high quality science illustrations in vector format easily searchable and filterable specifically for the life science community. All icons are under permissive CC0 or CC BY license. ➡️
Colab notebooks are great to share demos for bio or chemistry tools but it is hard to find them and they break easily so
@kmjablonka
and I built an aggregator website powered by Github Actions 1/
DiffDock is now on
@huggingface
Spaces with interactive visualization of the diffusion steps. Built using
@Gradio
and 3dmol.js and you can upload your own ligands as SMILES or sdf/mol2. Happy docking!
1/2
Excited to share DiffDock, new non-Euclidean diffusion for molecular docking! In PDBBind, standard benchmark, DiffDock outperforms by a huge margin (38% vs 23%) the previous state-of-the-art methods that were based on expensive search!
A thread! 👇
ProteinMPNN
@huggingface
now lets you predict all sequences in one go with AlphaFold2 and sort them by RMSD, score or pLDDT to pick interesting ones. For now you can run 15 seqs per run. 1/2
Sneak peek at the upcoming Bioicons
@inkscape
clipart extension. Simply search for illustrations directly within Inkscape (some other providers WikiCommons, Reactome are also available). A fresh batch of icons is also coming soon.
#creativecommons
#scicomm
In order to further the development of an open database of scientific illustrations for the life sciences we have now set up a non-profit under the umbrella of
@OC_Europe
1/
ProteinMPNN + ESMFold now available on
@huggingface
Spaces. Any monomeric protein < 400 aa can be redesigned instantly and structures are predicted immediately using ESMFold API. 1/3
We investigated AlphaFold3 against RoseTTAfold-AllAtom and a new variant of Metal3D for prediction of metal-protein complexes. Tl;dr AlphaFold3 is as good as specialized tools on metal ion prediction. RFAA not so much.
Some interesting findings in below thread: 1/
Two years ago BioIcons was launched and the database has grown by almost 800 icons thanks to many scientists contributing illustrations ☺️
Now we're looking to grow more and need your feedback and support... 1/2
I quickly put together a
@matplotlib
version of this using
@RDKit_org
. Simply enter a SMILES string and your values as list and position the pie chart :)
Prints out an SVG which you can edit in Inkscape or Illustrator.
@nathandao1109
Thanks! We like to use this type of figure to concisely illustrate substrate scope. Figure guru
@ScienceBey
was the person in the group to pioneer this type of figure and it has spread like wildfire🔥
Use RoseTTAfold2 directly from PyMol with this
@gradio
demo:
Interoperability is a key aspect of FAIR code. Glad that
@Gradio
makes it a piece of cake to build demos that can be interfaced with other programs via an API. 1/2
New version of the automatic radial scope plot, now with easier setup, automated positioning of the plots, organic substituents etc. Thanks to
@dr_greg_landrum
for the idea on how to automatically position the pie charts.
cc
@NarayanLab
@ScienceBey
Happy to present our new preprint on a deep-learning based method for rapid localization of transition metal ions in protein structures: Metal3D. Work done with
@andrea_levy
and
@lcbc_epfl
1/6
Many researchers including myself assumed that a
@biorender
publication license meant one could publish in any kind of journal including open access journals. However, the academic license forbids publishing CC BY implicitly. So how many papers are affected? 1/
You can now use Bioicons (and other free science illustration libraries such as
@reactome
) directly from within Inkscape 1.2 if you click on File ➡️ Import Web Image ➡️ Select Bioicons in Dropdown. 🥳🥳 1/2
New feature for proteinMPNN
@huggingface
- Positions can now be fixed with simple selection language such as "within 5 of resname ZN" or "within 6 of nucleic" to generate diverse sequences for enzymes or ligand binding proteins. 1/2
New
@gradio
custom component for 2D molecule input using Smiles strings. Molecules can be input as strings or drawn using Ketcher.
Demo and install instructions here
1/2
Let's have more virtual conferences even after travel restrictions are lifted. They are good for the environment and work-life balance. In order to help, I am designing a web platform that facilitates setting up a virtual meeting with a virtual poster session. 1/6
First beta release of a native Molecule viewer for Gradio built on 3dmol.js: Gradio Molecule3D
Simply install it using pip install gradio_molecule3d 1/3
If you use a webkit browser you can directly copy the icons as SVG into the clipboard and paste them directly in Illustrator and change colors, delete parts you don't want etc 🔥 (In Inkscape you currently get an image, but support for svg pasting is coming soon)
THIS IS BIG! 👀
It's now possible to take any of the >30,000 ML apps from Spaces and run them locally (or on your own infrastructure) with the new "Run with
@Docker
" feature. 🔥🐳
See an app you like? Run it yourself in just 2 clicks🤯
In addition to *fold, there are lots of structure➡️sequence models out there for "inverse folding".
@ginaelnesr
and I tried to collect them here: Let us know if we missed any!
Interesting new method for metal protein docking using a U-Net to predict a distance matrix of where the metal binds.
Here is an unsolicited peer review of this work: (Also posted on BioRxiv Comments awaiting moderation)
This week, we'll be having an interactive workshop with
@simonduerr
on Tuesday (3/28) @ 4 pm EST!
Simon will give an overview + live demos on tools to publish and deploy models (like AF2, ProteinMPNN etc.) as web apps.
See more at:
Happy to present our new preprint on a deep-learning based method for rapid localization of transition metal ions in protein structures: Metal3D. Work done with
@andrea_levy
and
@lcbc_epfl
1/6
Looks like there'll be no open code for AlphaFold3. At least, there is a server with 10 jobs per day and a server UI that looks pretty much like the
@Gradio
custom component I built for RoseTTAfold-AllAtom
#opensource
We have also launched AlphaFold Server, a free platform that scientists around the world can use for non-commercial research. 🔬
They can harness AlphaFold 3’s predictions and test hypotheses with just a few clicks - no matter their technical expertise. →
It's nice to see people questioning BioRender. While they do not claim copyright over the complete figures, scientists never own the figures fully if they use any of the icons therefore publishing the figures in open access journals under CC BY isn't possible! 1/
So many cool Google Colab demos these days on Twitter in the chemistry/protein space. Twitter search is pretty bad to find these again after some time. I wonder if the community should provide a space to collect links to all these demos for further inspiration. 1/2
1/5 Robust deep learning based protein sequence design using ProteinMPNN The high rate of experimental success and applicability to almost any protein sequence design problem has made it the standard approach at the
@UWproteindesign
Results for this competition are online since a few days. Interesting that the very good intermediate results (Rs=0.8+) were mainly due to people overfitting to the public test data 🥸. Final winner achieved Rs ~0.55 for the private test data. 1/2
Looking to organize a virtual conference with Poster session?. Try the beta version of my Virtual Conference Management System (VCMS) here First release will be soon, after weeding out some further bugs. List of features ⬇️
Need an icon for your favorite type of vaccine? Or perhaps an NMR spectrometer? Or for your next cool advance in protein ML modeling? If you want to contribute your own illustrations simply fork the repo and submit a pull request as described in the Readme
Scrolling through the hilarious "I don't like jupyter notebooks talk" by
@joelgrus
- found this "But imagine" slide and looks looks like a pretty good description of what
@Gradio
does (except for 2)
@baym
@wc_ratcliff
@BioRender
@cshperspectives
I have explicit confirmation from Biorender that CC BY was not okay without explicit written permission obtained by writing to a specific non public email address. As of November 2023 no CC BY at all possible anymore.
Interesting complement to our analysis on AF3 for protein-metal interactions. While for metals local context is used (e.g for His->Ala mutations), for other organic ligands like ATP AF3 does ignore local context of the pocket (e.g mutating all pocket residues to Phe).
Now also available as vector art on under Creative Commons 0 license. Thanks
@sasha_sundstrom
You also are as talented as Sasha and have illustrations to share with the community? Submit them on
I drew some instruments for a presentation and then decided you should ALL have these. You can have these as nice clean TIFF files (also pasted into a PPT). Please enjoy and use for whatever!
#RealTimeChem
#ChemTwitter
Specifying the input for cofolding models just got easier with my new gradio component "gradio_cofoldinginput".
Try it here to generate inputs for RoseTTAfold2 All Atom predictions including covalent modifications:
@Dani_M_Stevens
Very cool! Do you know about ? 1700 vector illustrations to kick-start Inkscape figures with some quality illustrations. I wanted to redraw a figure from a paper using Inkscape+BioIcons and record it for YT. Happy to contribute it to your platform.
If you work with Jupyter Notebooks I recommend you to ditch the web interface and switch to VSCode. Such a nice experience with proper code completion and configurable theming.
We tried to solubilize the claudin, rhomboid, and GPCR membrane folds. Using vanilla ProteinMPNN, we were not able to design soluble versions of these proteins, so
@JustasDauparas
trained a new version called SolubleMPNN resulted in many soluble designs of these folds.
A good reminder to never only make available research code on GitHub but to keep an archived copy for releases on Zenodo. It's just a few quick and easy steps:
Not so fun news.
@github
made an unilateral decision to block not only the Making-it-Rain repo but they also blocked the entire account of our main dev
@pablitoarantes
without any explanation.
Any explanation,
@github
?
Pretty cool consequence of AF3 using diffusion is that pLDDT is no longer just for C-alpha but for each atom. RFAA still uses C-alpha pLDDT. Should enable interesting applications for locally refining rotameric states of predictions.
@GeneCollector
@drawbiomed
Thanks for highlighting! Actually if you open a recent Inkscape version you can click on Import Web Image in the File menu and select BioIcons in the drop-down 😊
68 new icons now on - including new Cell culture and molecular modelling categories. Thanks
@MarcelTisch
for contributing! If you want to share your own creations please do get in touch.
Curious about the scientific background behind ProteinMPNN? Tune in to this talk 4pm EST/1pmPST/10pmCEST. While you wait for the talk you can play with the interactive demo on
@huggingface
:
24 hours until
@JustasDauparas
's talk on designing proteins with ProteinMPNN!
Sign up here for updates and zoom links:
DM day of if you need the link!
@AdrianoAguzzi
Very true! Unfortunately good illustrations also take time and skill. I tried to get funding to make BioIcons bigger but so far no luck 😢
💡 Use to cite references on slides instead of Lastname et al., Journal, 2022. Works for any long DOI from a journal, BioRxiv, Zenodo or DataCite.
#ux
#AcademicChatter
We will build Model Gardens; i.e., collections of curated models linked with data/computing resources needed to advance research communities. Gardens will include automated model software/hardware/task testing and uncertainty quantification.
Thanks to
@sokrypton
's AF2 speed trick inference for 6MRR now takes mere 20 seconds right in your browser without install or any other waiting time. Thanks
@huggingface
/
@Gradio
for making this incredibly easy. 2/3
Added over 400 new illustrations from
@togopic
to bioicons (e.g LC-MS) Thank you/ありがとうございます
@dbcls
AND you can now more easily submit your own illustrations. 1/2
This is common pattern also for
@NaturePortfolio
. Some examples: I tried requesting data from a paper "available on reasonable request". When the authors didn't respond I contacted the journal as lined out in their open data guidelines. No response since last october! 3/
3/ Collectively, scientists spend probably 17.9 - 33.8 Mio $ per year on BioRender. I've pitched an idea for an open database to many different funders but so far no one has taken me up on it. With a tenth of BioRender revenue we could establish a forever free database. 4/4
Bioicons ❤️
@drawio
/
@inkscape
- Two new extensions for Bioicons are now available: 1) -Go to , select the categories you need and launch the editor. Save locally or in the cloud and embed the drawings in your slides.
BioIcons has reached 1000 stars on github 🤩
The project is looking for new developers and submissions for new illustrations are always welcome on Reach out if you're interested in contributing.
Really glad to spend an extended summer in the Bay Area as a visiting researcher in Possu Huang's lab. Thanks
@firmenich
/
@EPFL_ReO
for financial support. 1/2
@baym
@DaniBeckman
@dava_keavney
@nquinoneso
It's even worse. In an Open Access journal like Nature Comms you actually cannot use BioRender figures since Biorender doesn't allow you to publish the figures you created under CC BY even if you pay for them.
It's unlikely this paper even needed this kind of peer review since a test of time by the community would've evinced any shortcomings anyway.
Nature clearly only cares about profit and reputation like all other big commercial publishers and not about open science. 2/
Props to the authors (
@GabriCorso
@HannesStaerk
et al.) that made the code available in contrast to most other papers for diffusion models in the bio space . Would love to add other models in the space as comparison. 2/2
Microsoft has a cool Google scholar alternative ready. Looks neat and can open links to papers in new tabs (which is a constant annoyance for me when scrolling through Google scholar profiles)
Kudos to
@PacBio
for releasing illustrations of their devices on as high quality vector art under CC0 for scientists to reuse for illustrating lab protocols or scientific articles ! More device manufacturers should be doing this! Thx
@jessemellon
PowerPoint has a pretty nifty presentation coach for rehearsing presentations. Tracks pace, voice pitch, originality and even highlights profane language. Already available on the web and apparently coming to Desktop soon. I'll be using this to rehearse for my next talk.
@sasha_sundstrom
@ReidAlderson
These are great ! I'm the developer of I'd be happy to vectorize them for you and put them on the platform if you let me know which license you want to use. Sounds like you would be okay with CC0, but CC BY with mandatory attribution also is a good choice
Looking for a biologist that knows the tree of life very well to help me categorize this nice collection of vector illustrations by
@pjkeelinglab
for 🧬
Please fill the BioIcons User Survey at and sign the Statement of Support. More Users/Supporters will make it will be easier to convince funding agencies to fund new illustrations. 2/2
Our Genetic Algorithm approach to predict protein thermostability for
#AIchem20posters
@RSC_CICAG
.
Explore the poster and some of the data interactively here: Preprint and next iteration using deep learning is in the works.
@baym
@wc_ratcliff
@BioRender
@cshperspectives
I also sent them two example articles with Biorender figures under CC BY and they told me the articles will need to be corrected. When I sent them 9000 more articles they ghosted me...
Google
@googlechrome
has tab groups now. Very helpful to structure the myriad of open browser tabs I always seem to accumulate :/ . Activate using chrome://flags "Tab Group" -> Enable
Also copyright violations: I sent
@NaturePortfolio
6454 papers with likely copyright violations in open access papers in January, they thanked me for the email but they still continue to happily publish the exact same copyright violations 4/
Want bigger proteins or multimeric ones? Clone the repo and run it locally or on colab with the same UI. You only need a somewhat beefy gpu with > 16GB RAM.
Instructions for local install here: 2/3
@BWusstsein
@Needie_
@Topfritte
Diese Rate gibt es nur in einer Provinz und nur für die Erstimpfung. Für wirksamen Impfschutz und Herdenimmunität braucht man allerdings zwei Impfdosen.
💡 Use to cite references on slides instead of Lastname et al., Journal, 2022. Works for any long DOI from a journal, BioRxiv, Zenodo or DataCite.
#ux
#AcademicChatter
"93% of the papers published in the first ten months of 2019 in Nature Materials declare that at least some of the data are available only from authors upon request..." What an argument for banning this phrase from any serious publication.
P.S Interactive manuscript is powered by
#manubot
/pandoc (DOCX, TEX->PDF and web version all generated from the same source) and web app can be easily implemented via
@huggingface
/
@gradio
/3dmoljs. Happy to help/provide advice if you want to host your own interactive manuscript.
A beautiful example is the MID1 protein (PDB 3V1C). Keeping the zinc binding residues fixed, ProteinMPNN is able to generate diverse sequences that fold back into the 3D structure with RMSD ~2-3 Å 2/2