Lucy Vost (@LucyVost) header
Lucy Vost (@LucyVost) profile picture

PhD student @OPIGlets working on ML methods for small molecule drug design mostly found @lucyvost.bsky.social

Oxford
Joined May 2022
Really happy that our work on conditioning diffusion models is published - have a read if you fancy hearing how we tried to encourage models to generate more physically plausible compounds by training conditionally with synthetic, distorted molecules!
In this week's first featured article Deane et al. introduce an approach for improving generative molecular design. Distorted molecules with quality labels are added to training data, allowing the model to distinguish unfavourable conformations. Read more: doi.org/10.1039/D4DD00331D
1
12
43
3,657
presenting this work at @NeurIPSConf AIDrugX and @workshopmlsb today - come by if you want to see how diffusion models for 3D molecule generation can be made a little more reliable! preprint: doi.org/10.1101/2024.09.17.6…
Replying to @OPIGlets
@LucyVost tells us how we can improve chemical plausibility in 3D molecule generation via conditional training with distorted molecules (preprint coming soon!)
1
2
211
The @OPIGlets are out in force at  @NeurIPSConf - check out our work at the @workshopmlsb and AIDrugX workshops today. @lucyvost, @nelequast and Nikhil Branson are presenting work on diffusion for small molecules, protein flexibility estimation and deep drug response models!
1
2
12
996
@LucyVost tells us how we can improve chemical plausibility in 3D molecule generation via conditional training with distorted molecules (preprint coming soon!)
1
2
1
3
741
OPIGlets are out in force at #AIChem24, organised by our very own @gmm DPhil students @LucyVost, @arun_raja007 and @isakvals are presenting posters, and Yael is giving a talk about her recent work MolSnapper on Wednesday! Please come and chat to them!
1
7
10
1,248
OPIG DPhil @YaeliZiv introduces MolSnapper: by integrating expert knowledge into diffusion models for structure-based drug design, we generate similar ligands to the ground truth, while doubling valid molecule output compared to alternative methods. biorxiv.org/content/10.1101/…
6
2
25
2,650
Happy to share our recent work on antibody library design, affinity prediction, and optimisation - 'Baselining the Buzz. Trastuzumab-HER2 affinity, and beyond!' Preprint (inc. SI) - doi.org/10.1101/2024.03.26.5… Code - github.com/oxpig/Tz_her2_aff… Data - doi.org/10.5281/zenodo.10549… 🧵
5
17
3
88
15,157
It took a while but it's here: the Fragmenstein preprint! chemrxiv.org/engage/chemrxiv… A big thank-you to everyone who helped in its creation, especially user feedback!
10
2
42
6,208
Charlotte's talk at the EMBO Workshop Computational Structural Biology (@embl) last December has just been released on YouTube: youtube.com/watch?v=2q18V57E…
1
4
52
4,128
The @covid_moonshot paper is out in Science: but can it help you as a compchem algo testset? Yes! All steps (frag to lead) crystallised & assayed, incl. negatives. Diverse methods, 4 series. Don't take my word, have a gander yourself: fragalysis.diamond.ac.uk/vie…
5
34
3,633