Sign in

Nicholas Runcie

@nicholasruncie.bsky.social
34 followers 31 following 9 posts

AI for drug design | Oxford DPhil student | MChem | RSci | opinions my own

PostsRepliesMedia
Nicholas Runcie @nicholasruncie.bsky.social · 13/05/2025
📝 Read the preprint “Assessing the Chemical Intelligence of Large Language Models” 👉 arxiv.org/abs/2505.07735
arxiv.org
Assessing the Chemical Intelligence of Large Language Models
Large Language Models are versatile, general-purpose tools with a wide range of applications. Recently, the advent of "reasoning models" has led to substantial improvements in their abilities in advan...
000
Nicholas Runcie @nicholasruncie.bsky.social · 13/05/2025
🪜 We believe these results signify a step change in the chemistry capabilities of LLMs. While current models are far from perfect, these models may soon, if not already, be able to solve a variety of problems in chemistry that were previously considered intractable (7/🧵)
100
Nicholas Runcie @nicholasruncie.bsky.social · 13/05/2025
However, these models are still far from perfect. 🔢 Counting atoms is tricky for an LLM, equivalent to counting the “r”s in strawberry. 🌀 Models struggled when representing molecules in a non-canonical format 📉 Performance was better for molecules with fewer atoms (6/🧵)
100
Nicholas Runcie @nicholasruncie.bsky.social · 13/05/2025
ChemIQ more closely resembles real-world challenges, with tasks including: 📈 SAR analysis ⚗️ Reaction prediction 🚀 NMR elucidation For example, o3-mini-high solved 74% of NMR spectra for molecules up to 10 atoms, and even solved one with 21 atoms. (5/🧵)
100
Nicholas Runcie @nicholasruncie.bsky.social · 13/05/2025
💥 SMILES to IUPAC conversion has been notoriously challenging, with all previous LLMs scoring ~0% on this task. In our evaluation, o3-mini-high named ~30% of structures correctly. (4/🧵)
100
Nicholas Runcie @nicholasruncie.bsky.social · 13/05/2025
💡Our results suggest that reasoning models now have the capacity to “think” about the structure of molecules, as opposed to talking about them superficially, enabling this step change in capabilities. (3/🧵)
100
Nicholas Runcie @nicholasruncie.bsky.social · 13/05/2025
🧪 We built ChemIQ as a benchmark for molecular comprehension and reasoning. o3-mini-high answered 59% of questions correctly, substantially improving on GPT-4o’s 7%. Additionally, more reasoning consistently improved performance across all questions. (2/🧵)
100
Nicholas Runcie @nicholasruncie.bsky.social · 13/05/2025
🚀 LLMs can now do chemistry! Our new preprint shows that state-of-the-art reasoning models can now perform advanced chemical reasoning tasks, without any assistance from external tools. Here’s what o3-mini can already do👇 (1/🧵)
132
Reposted by Nicholas Runcie
Achira @achira-ai.bsky.social · 21/02/2025
We're excited to announce we've raised a Seed round with Dimension, Amplify, NVIDIA NVentures, and Compound! Read more about our mission at achira.ai
085
Reposted by Nicholas Runcie
Odysseas Vavourakis @odyv.bsky.social · 27/01/2025
It’s an exciting time in protein design! 🧬✨ But much of the therapeutic potential—especially for antibodies—remains untapped. Why? 🤔 Antibodies seem like ideal candidates for design! 💉 Here’s a quick thread summarising our new review paper on the state of antibody structure prediction. 👇 1/
sciencedirect.com
Challenges and compromises: Predicting unbound antibody structures with deep learning
Therapeutic antibodies are manufactured, stored and administered in the free state; this makes understanding the unbound form key to designing and imp…
181
Reposted by Nicholas Runcie
John Chodera @jchodera.bsky.social · 17/01/2025
Incredibly excited to see how well ADMET, affinity, and pose prediction models *actually* perform in the ASAP x OpenADMET x PolarisHub blind challenge: polarishub.io/competitions
polarishub.io
Polaris
The benchmarking platform for drug discovery
0163
Reposted by Nicholas Runcie
Oxford Protein Informatics Group (OPIG) @opig.stats.ox.ac.uk · 17/01/2025
New blog post from 1st year DPhil student @nicholasruncie.bsky.social on visualising cheminformatics data as molecular networks
021
Nicholas Runcie @nicholasruncie.bsky.social · 17/01/2025
🔥 Molecule networks 🔥 New cheminformatics data visualisation: 📊 Rapid interpretation of SAR 🔊 Find molecules in the noise ❌ No more lists of molecules My new favourite visualisation method! 🚀 Read how I visualise Free Wilson analysis here www.blopig.com/blog/2025/01...
041
Reposted by Nicholas Runcie
Matteo Ferla @matteoferla.bsky.social · 14/01/2025
Fragmenstein is finally published! ⌬🔩📗 This tool stitches together the atoms of parent hits and re-animates (minimises) them into a conformation close to the parents, preserving their interactions: a must for crystallography-driven drug discovery doi.org/10.1186/s133... #chemsky #drugdiscovery
doi.org
Fragmenstein: predicting protein–ligand structures of compounds derived from known crystallographic fragment hits using a strict conserved-binding–based methodology - Journal of Cheminformatics
Current strategies centred on either merging or linking initial hits from fragment-based drug design (FBDD) crystallographic screens generally do not fully leaverage 3D structural information. We show...
043