Yew Mun

@AscentBio Great, I would like to try it out! I just tried it and one of the steps gave: Execution error: Sandbox not available

We are excited to announce that Faraday, our AI scientist for drug design, is now available in public beta. Agentic reasoning. Frontier Models. Seamless code execution. Starting today, researchers and drug developers worldwide can register and begin using Faraday to accelerate their discovery workflows. Public beta is now open at platform.ascentbio.ai

@kevinlu625 I would love to have a go at using the IDE!

The world's first vibe coding IDE A true next step forward in building with AI Comment for 100k free credits!

@theboymanager @StJsn1 .

@svpino I agree with you, so how can we quantify the amount of “thinking” in a business/career perspective?

Today, more people than ever can write code. But the number of people who can take a problem, analyze it, break it down, and write good, working code hasn't increased much. We are automating *typing*, but we can't automate *thinking*.

🚨preprint🚨 Read our new work on reactive metallo-scaffolds (r-mS)! We combine metal complexes with chemoproteomics to discover new liganding sites across the proteome! Shout out to a great collaboration with GSK! Fantastic team effort 🫶🏻chemrxiv.org/engage/chemrxi…

EMBL’s AI initiative strives to transform all areas of life science research through full AI integration. Major funding from the Hector Foundation is dedicated to attracting interdisciplinary talent and boosting infrastructure, training, & innovation. embl.org/news/connectio…

📣Preprint 📣 You want to degrade a protein in Gram-negatives? We have tackled this challenge! Detailed study of our Light-Activated Metal-Dependent Protein Degradation LAMP-D system in targeting and rescuing NDM-1 activity. What an amazing team effort 🫶🏻 doi.org/10.26434/chemr…

Join our lab, as a S/PLRS @TheCrick /KCL -Synth. Chemist/Chem. Biologist #chemtwitter #chemistry, to lead projects, build teams, support and train, collaborating across my group and more widely, driving forward basic+translational science tinyurl.com/5n8ek33x closes 25/6! 24mo

I'm using the Fellou AI browser to publish this post.

Small proteins can be more complex than they look! We know proteins fluctuate between different conformations- but how much? How does it vary protein to protein? Can highly stable domains have low stability segments? @ajrferrari experimentally tested >5,000 domains to find out!

🚨Less than a week left to apply (deadline 19/03/25) for this exciting fully funded 4 Year Phd position in my group. Multidisciplinary project on the design of novel metallo-NBTIs including all the fun of chemistry and much more! Please RT! Apply here: crick.ac.uk/careers-study/…

@BiologyAIDaily I think the GH link to the code is broken.

HybridLinker: Topology-Guided Posterior Sampling for Enhanced Diversity and Validity in 3D Molecular Linker Generation 1/ HybridLinker is a novel framework for 3D molecular linker generation, designed to overcome the trade-off between diversity and validity in fragment-based drug discovery. It integrates Point Cloud-Free (PC-Free) and Point Cloud-Aware (PC-Aware) models to achieve both diverse and valid molecular linkers. 2/ Traditional PC-Free models prioritize diversity but often generate invalid structures, while PC-Aware models ensure spatial validity but suffer from low diversity. HybridLinker bridges these two approaches using a novel two-step generation pipeline. 3/ At the core of HybridLinker is LinkerDPS, the first diffusion posterior sampling (DPS) method tailored for molecular generation. It guides a PC-Aware model with diverse bonding topologies sampled from a PC-Free model, refining the molecules to ensure both structural validity and exploration of novel topologies. 4/ HybridLinker significantly outperforms existing linker generation methods in both diversity and validity metrics. In evaluations using the ZINC dataset, it achieved a balance between generating novel molecular linkers while maintaining chemical feasibility. 5/ The framework enables zero-shot inference, meaning it does not require additional training. Instead, it leverages pretrained PC-Free and PC-Aware models, making it highly adaptable to advancements in generative molecular modeling. 6/ HybridLinker’s ability to balance diversity and validity makes it particularly useful for complex molecular design tasks such as PROTAC linker generation, scaffold hopping, and fragment-based drug discovery, where maintaining structural integrity is crucial. 7/ Beyond linker generation, the introduction of LinkerDPS has broader implications for molecular design. It establishes a new paradigm for integrating topological and spatial information in generative models, which could be extended to protein design and structure-based drug discovery. 💻Code: github.com/HybridLinker/H… 📜Paper: arxiv.org/abs/2502.17349 #DrugDiscovery #AI #MachineLearning #MolecularDesign #FragmentBasedDrugDiscovery #DiffusionModels #Bioinformatics #ComputationalChemistry

@janhjensen @MariaHarrisRas1 @Strandgaard96 @Angelo_Frei @BalcellsD This is very useful!

New preprint: SMILES All Around: Structure to SMILES conversion for Transition Metal Complexes doi.org/10.26434/chemr… #compchem with @MariaHarrisRas1 @Strandgaard96 Julius Seumer Laura Hemmingsen @Angelo_Frei @BalcellsD

@BiologyAIDaily I don't think the link to the code is available at the moment.

Autoparty: Machine Learning-Guided Visual Inspection of Molecular Docking Results 1. Introducing Autoparty, an innovative Python-based tool designed to accelerate molecular docking analysis through active learning and human-in-the-loop annotation, significantly enhancing hit identification in drug discovery. 2. Autoparty leverages interface fingerprints (IFPs) to represent protein-ligand interactions, providing accurate insights into docking poses and their biochemical relevance, surpassing traditional scoring functions. 3. The active learning framework queries users for the most informative molecule annotations, building highly accurate machine learning models with minimal labeled data. 4. In a real-world case study, Autoparty doubled the hit rate among 193 experimentally tested compounds compared to standard docking score-based selection methods. 5. The tool supports interactive visualization using 3Dmol.js, enabling medicinal chemists to efficiently evaluate intermolecular interactions and prioritize compounds. 6. Autoparty’s modular backend allows seamless integration of custom acquisition functions, annotation strategies, and uncertainty quantification methods to suit diverse screening needs. 7. The system supports annotation persistence with an SQL database, facilitating reproducibility and further downstream analyses in virtual screening campaigns. 8. Designed with accessibility in mind, the tool is containerized for portability and ease of deployment across computational environments. @LauraShub 💻Code: github.com/keiserlab/auto… 📜Paper: doi.org/10.26434/chemr… #DrugDiscovery #MachineLearning #VirtualScreening #Bioinformatics #ActiveLearning

Super excited to preprint our work on developing a Biomolecular Emulator (BioEmu): Scalable emulation of protein equilibrium ensembles with generative deep learning from @MSFTResearch AI for Science. #ML #AI #NeuralNetworks #Biology #AI4Science biorxiv.org/content/10.110…

📣Reminder📣 AI-First Drug Design: Accelerating the Discovery of New Therapeutics | drughunters.com/3VoceVz Thu, Dec 12, 2024 8 AM PST / 11 AM EST / 5 PM CEST Rebecca Paul will share her insights on how AlphaFold 3 and other breakthrough AI models are applied in drug discovery.

Today we’re previewing NeuralPLexer3 Beta: demonstrating unprecedented accuracy and speed in predicting protein-drug complexes, with instant and accurate structural insights across the full range of protein classes and drug molecules. Learn more: iambic.ai/post/np3-previ…

Exciting opportunity🚨 Joint PhD position with @lab_mueller 👇🏻Application deadline 29th October 24! kcl-mrcdtp.com/project/ration… Retweets much appreciated!

Doing good science is 90% finding a science buddy to constantly talk to about the project.

Congratulations to Simon Boulton, one of our assistant research directors, on his recent appointment as @CR_UK’s new chief translation advisor 🥳