An open source toolkit, developed by and for the machine learning community, that simplifies molecular processing and featurization workflows.
Joined January 2021
- Tweets142
- Following506
- Followers918
- Likes442
Pinned Tweet
Introducing Medchem, an open source Python library for molecular filtering and prioritization.
It includes hundreds of molecular filters, alerts, and rules, both established and novel.
Read the blog by @HadiM_ for more details: portal.valencelabs.com/blogs…
Spotted some molfeat users at the @Polaris_HQ hackathon 👀
They're searching for quick and easy ways to test different molecular featurizers.
molfeat.datamol.io/featurize…
5
287
Catch @ENoutahi @craigmichaelm @TOSSOUPrudencio @jonathanlimsc at #NeurIPS tomorrow!
They'll be presenting the SAFE paper at two workshops between 12:00 and 2:30 CST.
See below for more details.👇
1⃣AI for Accelerated Materials Design
Where: Room 228 -230
When: Poster session from 12:00 - 2:00 pm CST on Friday, December 15th
sites.google.com/view/ai4mat…
1
92
Looking for events that bring the TechBio community together?
Check out the mixers happening at Oxford and Cambridge next week.
We’re hosting two TechBio mixer events next week! One co-hosted with @mmbronstein in Oxford, and one hosted on campus @cambridge_uni.
Join us for an evening of dinner, drinks, and discussions on accelerating AI with drug discovery.
RSVP through the event pages below.👇
2
404
Today, we are excited to introduce Graphium into the datamol.io ecosystem!
Graphium is a powerful and flexible Python library for training molecular GNNs at scale.🚀♾️
Try it out today: github.com/datamol-io/graphi…
1/ Can ChatGPT generate molecular featurizers? Do we need to finetune LLMs on molecular context to get good molecular representations? 🤔
We tried this with the molfeat-hype extension using embeddings provided by various LLMs 👇
There are waaay too many molecular featurizers out there - here’s a breakdown by category of what’s available on molfeat.
We suggest trying out the pre-trained models first! 🧠 Which featurizer does the community want to see next in molfeat?
Let us know👇
molfeat makes it easy for ML scientists working in drug discovery to evaluate and implement a wide range of featurizers, see what’s available here: molfeat.datamol.io/featurize…
Find us on GitHub: github.com/datamol-io/molfea…
5
287
1/ Shoutout to datamol contributor Desmond Gilmour for introducing lasso highlighting! 🪢
This new feature enhances the visualization process on datamol, allowing users to quickly identify and visualize specific parts or features of a molecule.
Try now: dev.to/dessygil/lasso-highli…
Transforming units, adding molecules and descriptors to Pandas dataframes, clustering, and filtering based on descriptor values. 🧪
Learn about how datamol can help you accomplish these tasks and a lot more in this awesome tutorial by @wpwalters!
A new Practical Cheminformatics tutorial highlights some of the features in the datamol package that simplify cheminformatics workflows. “Streamlining Cheminformatics Workflows with datamol 1. Data manipulation, descriptors, and clustering”.
colab.research.google.com/gi…
0/ 🧩 We’ve added support for PaDEL descriptors to molfeat!
The first molfeat plugin now supports the PaDEL molecule featurizers which have a total of 797 descriptors and 10 types of fingerprints. 🤯
Get started here: github.com/datamol-io/molfea…
Super exciting to see this from @wpwalters, we’re looking forward to future posts! 🚀
6
915
0/ Today, we’re excited to introduce molfeat, a new hub of molecular featurizers.
Easily discover and implement a diverse range of featurizers directly in your ML workflows.
Read more here: m2d2.io/blog/posts/introduci…
It’s always cool to see Datamol being used in a variety of research and papers. 😎
Awesome work @AdeleneLai!
🚨Our new paper in @jcheminf on automatic homologous series classification in compound datasets
jcheminf.biomedcentral.com/a…
Jonas Schaub @csteinbeck @ESchymanski @RDKit_org
#polymers #environmentalchemistry #PFAS #naturalproducts #cheminformatics #algorithm
5
399
Try the smaller, 11M parameter version, of the award-winning GNN model for molecular property prediction!
Available now on @HelloPaperspace: hubs.la/Q01tjkHG0
You can also sign up for early access to the full-size GNN model, MolGPS, here: datamol.io/molgps
What do you get when you take the best qualities of conventional #GNNs and merge them with transformer-style attention?
Meet GPS++, the award-winning #GNN model from Graphcore, @Mila_Quebec & @valence_ai.
Try for free now on @HelloPaperspace: hubs.la/Q01tjkHG0
Announcing the release of MolGPS, the largest pre-trained model for molecular featurization, built on top of the 2022 OBG winning architecture, GPS++, for state-of-the-art molecular property prediction.
Sign up for early access on Datamol here:
datamol.io/molgps
Read more about GPS++ and the submission to the 2022 OGB Large Scale Challenge here:
m2d2.io/blog/posts/valence-a…
2
In the past few threads, we’ve talked about how important structure-activity relationships (SAR) are for drug discovery.
Efficiently aligning the molecules in your dataset can make it easier to visualize and highlight substructures that are relevant to activity. 🧵(1/5)
Aligning molecules can also help visualize shared structures between compounds and identify potential activity patterns through activity cliffs. 🧵(2/5)
ncbi.nlm.nih.gov/pmc/article…
1
1
In Datamol, “fuzzy scaffolding” is a useful concept for constrained scaffold generation.
This feature enables finer control over the types of scaffolds you can generate as you can enforce specific groups to appear in the molecular core. 🧵(1/5)
Enforcing specific groups in the core also acts as a pharmacophore requirement.
Pharmacophores are defined as “a part of a molecular structure that is responsible for a particular biological or pharmacological interaction that it undergoes”. 🧵(2/5)
link.springer.com/referencew…
1
1
Scaffolds are best defined as molecular “cores” of a compound where you have functional groups attached via attachment points.
By identifying these core structures, we can leverage the structure-activity-relationship (SAR) of compounds to design new and better drugs. 🧵(1/9)
Although there are technically many definitions for a scaffold, the most common one is the Bemis-Murcko scaffold. 🧵(2/9)
pubs.acs.org/doi/10.1021/jm9…
1
7
What are some of the strategies that we can use to develop effective therapeutic agents?
One approach is using fragments as building blocks. These are small molecules with low molecular weights, often representing a small part of a bigger, drug-like compound. 🧵(1/11)
Fragments are useful as starting points in a medicinal chemistry campaign. They are used to identify small chemical and functional groups that bind, even if weakly, to the target of interest. 🧵(2/11)
en.wikipedia.org/wiki/Medici…
1
3
Thank you to @iwatobipen for writing a blog on how easy it is to use Datamol! He’s written countless cheminformatics blog posts which I’m sure many of us have read at some point when looking for an RDKit feature/tutorial.
Check it out here:
iwatobipen.wordpress.com/202…