Benoît Baillif

I am happy to share that our preprint on benchmarking structure-based 3D molecular generative models is online: arxiv.org/abs/2407.04424. Thanking my co-authors @AndreasBenderUK , @mccabe_p_ and @JasonColeCCDC 1/10

gênant (à 100 000 RT on le fait)

I’m grateful to have been awarded the Best Poster at EuroQSAR 2024 🎉🏆 for our work on Uncertainty Quantification in Bioactivity Assessment.

#GenBench3D: A game-changer for #drugdiscovery! 💊 Assess AI-designed molecules' true potential with this tool. TargetDiff and Pocket2Mol lead in Vina, Glide, and Gold PLP scores using valid relaxed molecules. Quick read: cbirt.net/evaluating-3d-… #bioinformatics #AI #sciencenews

2000 étudiant-es ont déménagé pour rien ? 🫠 Face à l’insalubrité des logements Crous la police se voit proposer un relogement en 24h ⤵️

💥 Introducing PINDER & PLINDER With existing evals saturating without clear advances in real life downstream tasks, current progress in AI x Bio is primarily rate-limited by better datasets & evals. In two back-to-back preprints, we address this via fantastic academic-industry collaborations by @vant_ai @NVIDIA & the group at @TorstenSchwede that co-organizes CASP, @MIT: Read the release blog posts at: plinder.sh/blog/plinder & pinder.sh/blog/pinder - 🧵(1/6)

Notre premier objectif (500 signatures !) a été atteint seulement deux jours après le lancement de notre appel. Nous en sommes très fiers ! ✊ Très bientôt, on vous dévoile le projet pour la suite. En attendant : les15engagementsnfp.fr

Interested in 3D Molecule Generation and Structure-Based Drug Design? Check out the demo of our Molcraft paper at #ICML2024 to generate, visualize, and evaluate ligands for different pockets! The demo is available at gensi-thuair.com:10990

🔴LFI suspend sa participation aux négociations NFP Notre communiqué ⬇️ Nous partageons la peur que beaucoup de personnes de gauche : la fin de la dynamique qui a porté la campagne, le retour des querelles, et à la fin, la victoire du bloc réactionnaire. les15engagementsnfp.fr

Looking forward to receive feedback on both the manuscript and the GenBench3D package!

We recommend to perform local minimization of the generated molecules in the binding pocket before structure-based assessment, and to use at least two scoring functions if you have access to (e.g. Glide or Gold PLP). 10/10

3D generative models producing molecules directly inside the binding pocket have the potential to find new binders by exploiting the pocket information with deep learning (see sciencedirect.com/science/articl… for a short review on the topic). 3/10

For this work, I developed the GenBench3D, publicly available on GitHub to test on your own generated molecules: github.com/bbaillif/genbe… 2/10

I am happy to share that our preprint on benchmarking structure-based 3D molecular generative models is online: arxiv.org/abs/2407.04424. Thanking my co-authors @AndreasBenderUK , @mccabe_p_ and @JasonColeCCDC 1/10

@wpwalters The preprint and code are coming soon! I presented an earlier version of this research at the Cambridge Cheminformatics meeting: youtu.be/uGkRmbJTQiI?si… and in a poster: figshare.com/articles/poste…

@wpwalters I focused on defining how likely ("valid") the conformations of generated molecules are, I came up with a solution to look-up bond length and angles distributions in the CSD.

New Practical Cheminformatics Post, "Generative Molecular Design Isn't As Easy As People Make It Look", practicalcheminformatics.blogspot.com/2024/05/genera…