@ian_dunn_ Profile picture

Ian Dunn

@ian_dunn_

PhD Candidate @CMUPittCompBio, advised by @david_koes. Interested in deep learning for molecular structures and accelerating drug discovery.

Similar User
Miruna Cretu photo

@MirunaCretu2

Oliver Schilter photo

@OSchilter

Wenhao Gao photo

@WenhaoGao1

Derek van Tilborg photo

@DerekvTilborg

BOjana Ranković photo

@6ojaHa

Matteo Manica photo

@drugilsberg

Lucky Pattanaik photo

@lucky_pattanaik

Benoît Baillif photo

@BaillifBen

Yuanqing Wang photo

@YuanqingWang

Jeff Guo photo

@JeffGuo__

Morgan Thomas photo

@MorganThomas263

Jannis Born photo

@JannisBorn

Ye Buehler photo

@Ye_Buehler

Adeesh Kolluru photo

@AdeeshKolluru

David Kreutter photo

@DavidKreutter

Pinned

New preprint! “Mixed Continuous and Categorical Flow Matching for 3D De Novo Molecule Generation” arxiv.org/abs/2404.19739… FlowMol is a flow matching model that jointly samples the geometric and topological structure of molecules. We also present surprising results on flow…


Ian Dunn Reposted

Finally, a community-wide paper @JCIM_JCTC led by @mattschap describing results of CACHE Challenge #1: Targeting the WDR Domain of LRRK2, A Parkinson’s Disease Associated Protein! #compchem #drugdiscovery pubs.acs.org/doi/10.1021/ac… Here we report the results of the inaugural…

Tweet Image 1

Ian Dunn Reposted

Our work on all-atom tokenization of any biomolecular structure is on arxiv now! arxiv.org/abs/2410.19110 @BiologyAIDaily summary is perfect, but to sum up the summary: - composability: small molecule, protein and RNA structure "point-clouds" can be described with the same vocab…

Bio2Token: All-atom tokenization of any biomolecular structure with Mamba @FlagshipPioneer • This paper introduces “Bio2Token”, a method that tokenizes biomolecular structures at an all-atom level using Mamba. Unlike many current approaches that rely on coarse-grained…

Tweet Image 1


Ian Dunn Reposted

Announcing CACHE2 results! Crowdsourcing and conventional methods did as well as novel ML. Teams started from 4 fragments in the PDB. We tested their predictions and found double-digit µM hits. This is the state-of-the-art. @thesgconline @consciencemeds bit.ly/4dTPBzO

Tweet Image 1

Ian Dunn Reposted

We’re presenting AlphaProteo: an AI system for designing novel proteins that bind more successfully to target molecules. 🧬 It could help scientists better understand how biological systems function, save time in research, advance drug design and more. 🧵 dpmd.ai/3XuMqbX


Ian Dunn Reposted

ESM3 is out from EvolutionaryScale! 98B params (~GPT3 scale). Multimodal over sequence, structure, and function with cool design applications. Trained on variable masking ratios and decodes proteins iteratively. Some work on alignment too. Looks exciting! evolutionaryscale.ai/blog/esm3-rele…


Ian Dunn Reposted

Great post on AlphaFold3 from @inductive_bio that I'm surprised has not got more attention🤔 - AF3 paper benchmarked Vina (not SOTA) using a single conf - Conventional docking has near AF3 performance when using multiple starting confs and Gnina scoring to select best pose (1/2)

Tweet Image 1

Ian Dunn Reposted

Excited to share our latest preprint: "Transferable Boltzmann Generators"! We propose a framework based on flow matching and demonstrate transferability on dipeptides. Work done with amazing @FrankNoeBerlin Check it out here: arxiv.org/abs/2406.14426


Ian Dunn Reposted

New blog post alert! 🚨 I write about AlphaFold3, how it works, where it works well ✅ (and where we think it might not ❌) and what it means to the TechBio landscape more broadly (if you decide to believe me). 🧬 (1/7) 🧵 open.substack.com/pub/harrisbio/…

Tweet Image 1

Ian Dunn Reposted

Introducing FoldFlow-2, a new SOTA sequence-conditioned protein generative model! work w/ @DreamFoldAI James V @FatrasKilian Eric TL @PabloLemosP @riashatislam @ChengHaoLiu1 @jarridrb @tara_aksa @mmbronstein @AlexanderTong7 @bose_joey Arxiv: arxiv.org/abs/2405.20313 1/8 🧵


Ian Dunn Reposted

Cool new paper with Julian Cremer, Tuan Le, Okko Clevert and @ktschuett, introducing a new way to do conditional generation through diffusion models via importance sampling of the denoising pathway. Applied to ligand generation in proteins. arxiv.org/abs/2405.14925


Ian Dunn Reposted

Wonder how to generate discrete data with Riemannian Flow Matching? Well, here is our most recent work addressing this issue: Fisher Flow Matching 😎 ➡️ Available here: arxiv.org/abs/2405.14664 🔥 Let me break it down 🧵 (1/10) @bose_joey @mmbronstein @ismaililkanc @MirceaSci

Tweet Image 2

Wonderful critical evaluation of a generative model for SBDD. Bad news: generated small molecules require heavy post-processing and many compounds are unstable. Good news: This provides clear goal posts for model performance that we should be working towards.

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

Tweet Image 1


Ian Dunn Reposted

I am happy to share our new ICML work, a new score-based generative model for molecules conditioned on protein structures. arxiv.org/abs/2405.03961 (below is an example of a single-chain generating ligands)


This raises many questions. Did anybody ever compare the performance of frame representations and equivariant architectures to simpler options? I think I’ve seen results suggesting equivariance helps. Why have researchers come to different conclusions about the utility of…

AF2: inspires a zillion diffusion/flow matching papers to use a frame representation of the backbone, AF3: diffusion over atomic coordinates with no equivariances built into the architecture or representation.



Ian Dunn Reposted

Very excited to present SynFlowNet next week at @gembioworkshop #ICLR2024! We tackle the synthesisability issue in molecular design by training a GFlowNet with an action space of chemical reactions. Thanks to my coauthors @charlieharris01 @juleroy13 @folinoid and Pietro Lio (1/n)

Tweet Image 1

Ian Dunn Reposted

We're excited to have @ian_dunn_ from the @david_koes come to the Rosetta Molecular ML Reading Group to present his work on "keypoint" latent molecular diffusion models Today at 11am EDT. See maomlab.github.io/mol_ml_reading… for the zoom link


Ian Dunn Reposted

David Koes and Ian Dunn: Docking with GNINA is all you need!

Tweet Image 1

Loading...

Something went wrong.


Something went wrong.