←
Dr Javier Sánchez Utgés @javierutges.bsky.social356 posts
Dr Javier Sánchez Utgés @javierutges.bsky.social avatar

Dr Javier Sánchez Utgés @javierutges.bsky.social

@JavierUtges

BSc in Genetics at @UABBarcelona MSc in Bioinformatics at @UPFBarcelona PhD in Structural Bioinformatics at @dundeeuni Postdoctoral Research Fellow at @ucl

◷ Joined July 2020▧ 82 Photos and videos
ISCB News avatar
⟳ Dr Javier Sánchez Utgés @javierutges.bsky.social retweeted
ISCB News@iscb·Oct 4

The scientific programme for 3D-BioInfo 2026 is live! 🧬And today is the last chance to take advantage of the discounted registration rate!  🔗Register: iscb.org/3dbioinfo2026/regis…

Post attachment
Dr Javier Sánchez Utgés @javierutges.bsky.social avatar
Dr Javier Sánchez Utgés @javierutges.bsky.social@JavierUtges·Oct 1

Are you working on the field of #Structural #Bioinformatics and have not registered yet? What are you waiting for? Join me and the community in attending this excellent meeting in beatiful and history-rich #Rome! I am so excited for this and cannot wait to see you all there!

🎉 Good news! The early registration deadline for the 3D  BioInfo annual meeting has been extended to Sunday, October 4!  Register at iscb.org/3dbioinfo2026/regis… 📌ELIXIR membership is NOT required.

Post attachment
Dr Javier Sánchez Utgés @javierutges.bsky.social avatar
Dr Javier Sánchez Utgés @javierutges.bsky.social@JavierUtges·Sep 21

5/6 🔹 For novel targets, DrugCLIP recovers the correct ligand within the top 1% of ~50,000 compounds in 55-75% of cases. But performance drops in more realistic scenarios, driven by sidechain orientation and, for predicted pockets, differences in pocket residue selection.

Dr Javier Sánchez Utgés @javierutges.bsky.social avatar
Dr Javier Sánchez Utgés @javierutges.bsky.social@JavierUtges·Sep 21

4/6 🔹 Ligand embeddings capture a more pocket-aware notion of chemical similarity, rather than simply reproducing fingerprint similarity. And after rigorous de-leakage, DrugCLIP still generalises to proteins AND chemistry unseen during training. 🧬💊

Dr Javier Sánchez Utgés @javierutges.bsky.social avatar
Dr Javier Sánchez Utgés @javierutges.bsky.social@JavierUtges·Sep 21

3/6 🔹 DrugCLIP sets a new state of the art for pocket similarity search – despite not being trained for this task – while running >100× faster than existing structural descriptors. Its pocket embeddings strongly track structural similarity and are robust across conformations.

Dr Javier Sánchez Utgés @javierutges.bsky.social avatar
Dr Javier Sánchez Utgés @javierutges.bsky.social@JavierUtges·Sep 16

Here are some of my favourite talks: - “Stoic: Fast and accurate protein stoichiometry prediction” by Daniil Litvinov - “Efficient sampling of large-scale transition pathways and intermediate conformations in sub-mesoscopic protein complexes” by Domenico Scaramozzino