Team-run account of the Molecular ML research group led by @fra_grisoni | #AI for Drug Discovery | Doing all this at @tueindhoven & @ICMStue
Eindhoven, Nederland
Joined September 2021
- Tweets207
- Following2.7K
- Followers3K
- Likes412
Pinned Tweet
👇
Happy two-year anniversary to the @molecularML team! 🎂
What a journey it has been so far — filled with immense satisfaction, cool collaborations, and incredibly talented people.
A toast to stretching the limits of #AI in drug discovery and chemical biology 🥂
Molecular Machine Learning retweeted
ELLIS ML4Molecules Workshop running at the Ellis Unconference at #EurIPS, Copenhagen.
Molecular Machine Learning retweeted
How can AI help us discover new molecules, materials, and medicines?
Join us for ML4Molecules 2025, part of the ELLIS UnConference in Copenhagen 🇩🇰 on Dec 2 co-located with #EurIPS
📣 Call for contributions open until Oct 15 (extended)! 🔗 More info: eurips.cc/ellis
Molecular Machine Learning retweeted
#DreamReactions2025: Congratulations to award winners! 🍃Green Dream Reactions Award🏆 to @fra_grisoni and poster awards to researchers from Mülheim, Münster, Würzburg and Amsterdam. Special thanks to Reinhard Hoffmann (center)!
Molecular Machine Learning retweeted
Scaffold Hopping with Generative Reinforcement Learning
pubs.acs.org/doi/10.1021/acs…
@foo_fighterin @fra_grisoni
#JCIM Vol65 Issue13 #MachineLearning #DeepLearning
Molecular Machine Learning retweeted
Molecular task arithmetic reduces off‑target drift:
It keeps other physicochemical properties closer to the pretrained baseline than standard finetuning
That subtlety matters for downstream ADMET
Molecular Machine Learning retweeted
Read the paper:
Look the Other Way: Designing 'Positive' Molecules with Negative Data via Task Arithmetic
arxiv.org/abs/2507.17876
Molecular Machine Learning retweeted
💥 Negative data ≠ useless noise for molecules
⚙️ Train on 10k negatives → walk another way → zero‑shot, dual‑objective +24k hit clusters with higher diversity than classic finetuning
Negatives-only Tradeoff: ‑14 % success‑rate & -16% validity → blend a few actives to rebalance 🧬 #AI4DrugDiscovery #MolecularTaskArithmetic @fra_grisoni @Rza_ozcelik @molecularML
New paper alert 👇
Our perspective on deep learning for de novo design is now published @JCIM_JCTC! We curated hundreds of works for beginners and experts, covering:
🔮 generative models
📝 synthesis planning
📊 benchmarks
🧪 translation to wet-lab
🎯 challenges ahead
Molecular Machine Learning retweeted
Our perspective on deep learning for de novo design is now published @JCIM_JCTC! We curated hundreds of works for beginners and experts, covering:
🔮 generative models
📝 synthesis planning
📊 benchmarks
🧪 translation to wet-lab
🎯 challenges ahead
Molecular Machine Learning retweeted
Our work on predicting protein binding sites with deep learning is now online! The work has multiple outputs:
Preprint: chemrxiv.org/engage/chemrxiv…
Web app of the model: 14-3-3-bindsite.streamlit.ap…
Python library for peptide processing: doi.org/10.1093/bioadv/vbaf0…
Let’s break it down! 🧵👇
New paper alert 👇
Our DeepCocrystal is now online @angew_chem!
DeepCocrystal is a deep learning model that predicts co-crystallization of small molecules from molecular strings. Convolutions shine again✨
AND, its predictions are tested in the lab! 🥼🧪
Molecular Machine Learning retweeted
Our DeepCocrystal is now online @angew_chem!
DeepCocrystal is a deep learning model that predicts co-crystallization of small molecules from molecular strings. Convolutions shine again✨
AND, its predictions are tested in the lab! 🥼🧪
Great to see our work out in @angew_chem! 🎉
We introduce ‘supramolecular’ language processing to predict co-crystallization — with experimental validation! 🙈
Led by the unstoppable Rebecca Birolo, w/ @Rza_ozcelik; @TUeindhoven, @molecularML; @unito 💪🏻
onlinelibrary.wiley.com/doi/…
Molecular Machine Learning retweeted
Great to see our work out in @angew_chem! 🎉
We introduce ‘supramolecular’ language processing to predict co-crystallization — with experimental validation! 🙈
Led by the unstoppable Rebecca Birolo, w/ @Rza_ozcelik; @TUeindhoven, @molecularML; @unito 💪🏻
onlinelibrary.wiley.com/doi/…
🔈 New preprint alert!
Proud of @hlnbrnkmnn’s first PhD paper on rethinking how SMILES augmentation is performed for generative #DeepLearning! 🚀
Check it out! 👇
📄 chemrxiv.org/engage/chemrxiv…
🖥️ github.com/molML/fantasticSM…
@molecularML @TUeindhoven
Molecular Machine Learning retweeted
Proud of @hlnbrnkmnn’s first PhD paper on rethinking how SMILES augmentation is performed for generative #DeepLearning! 🚀
Check it out! 👇
📄 chemrxiv.org/engage/chemrxiv…
🖥️ github.com/molML/fantasticSM…
@molecularML @TUeindhoven
Molecular Machine Learning retweeted
Are you using generative #DeepLearning for de novo molecule design?🧪 🖥️
Then check out @Rza_ozcelik ‘s latest work, where we perform a (super) large scale analysis (~1 B designs!) & find ‘traps’, ‘treasures’ and ‘ways out’ in the jungle of generative drug discovery.
🌴 🐒
👇
📢 "How to evaluate de novo designs?" 🤔 If you were ever intrigued by this question, we have a fresh preprint for you!
We dig into evaluating design libraries and discover pitfalls and solutions — all in a jungle analogy!
arxiv.org/abs/2501.05457
A thread 🧵
Molecular Machine Learning retweeted
"The Jungle of Generative Drug Discovery: Traps, Treasures, and Ways Out" by @Rza_ozcelik , @fra_grisoni
Paper: arxiv.org/abs/2501.05457
#machinelearning
Molecular Machine Learning retweeted
📢 "How to evaluate de novo designs?" 🤔 If you were ever intrigued by this question, we have a fresh preprint for you!
We dig into evaluating design libraries and discover pitfalls and solutions — all in a jungle analogy!
arxiv.org/abs/2501.05457
A thread 🧵
Preprint alert! ‼️
Are you using generative #DeepLearning for de novo molecule design?🧪 🖥️
Then check out @Rza_ozcelik ‘s latest work, where we perform a (super) large scale analysis (~1 B designs!) & find ‘traps’, ‘treasures’ and ‘ways out’ in the jungle of generative drug discovery.
🌴 🐒
👇
Molecular Machine Learning retweeted
If you are doing peptide informatics, check out our 'tidy and tiny' python package! Should be useful especially for new researchers to start building machine learning pipelines.
Code, preprint, and documentation are available 🙂
peptidy: A light-weight Python library for peptide representation in machine learning
- Researchers have introduced peptidy, a lightweight Python library that facilitates converting peptides into numerical representations suitable for machine learning. This tool could accelerate peptide discovery and optimization in various fields such as drug discovery and food technology.
- Key features of peptidy include: no external dependencies, seamless integration into modern Python environments, support for a range of encoding strategies suitable for both predictive and generative machine learning, and support for post-translational modifications like phosphorylation, acetylation, and methylation.
- The library implements five popular peptide encoding methods, including peptide descriptors, amino acid descriptors, BLOSUM62 encoding, one-hot encoding, and label encoding. All representations except global descriptors are also suitable for generative deep learning.
- The authors benchmarked peptidy representations using literature data to predict antimicrobial activity and toxicity. Results showed that different encoding strategies have varying performance depending on the task and metric, emphasizing the value of experimenting with alternative representations when training peptide machine learning models.
- Peptidy is accompanied by extensive documentation and tutorials to facilitate accessibility. It is also open-sourced to allow feedback from researchers and extend its capabilities. The package can be easily installed via pip, and the installation instructions, Python code, and benchmarking datasets are available on GitHub.
@fra_grisoni
💻Code: github.com/molML/peptidy
📜Paper: doi.org.org/10.26434/chemrxi… #peptidy
#MachineLearning #PeptideRepresentation #PythonLibrary #DrugDiscovery #FoodTechnology
