Matthieu Schapira
@mattschap.bsky.social
Prof @ U. Toronto - PI @ SGC - CACHE challenges - Computational chemistry- Structural bioinformatics - Biophysics
What if, instead of trying to predict properties of every molecule, we focus on simply ranking them? After all, when running Bayesian optimization (BO) for drug/materials discovery, what matters is picking the best candidates first. Paper: doi.org/10.1063/5.02... Code: github.com/gkwt/rbo [1/5]
CACHE 7 is launched with support from the @gatesfoundation.bsky.social and unpublished data from Damian Young at @bcmhouston.bsky.social, Tim Willson @thesgc.bsky.social and Neelagandan Kamaria InSTEM. Design selective PGK2 inhibitors. We'll test them experimentally. bit.ly/4lnVYOs
New @chemrxiv.bsky.social preprint! RoboChem-Flex is a powerful, low-cost (<5k EUR), modular self-driving lab for chemical synthesis We showcase 6 studies (photochemistry, biocatalysis, cross coupling, ee ...), all optimized with different configurations & ML 🔗 chemrxiv.org/engage/chemr...
1/4 🚀 Announcing the 2025 Protein Engineering Tournament. This year’s challenge: design PETase enzymes, which degrade the type of plastic in bottles. Can AI-guided protein design help solve the climate crisis? Let’s find out! ⬇️ #AIforBiology #ClimateTech #ProteinEngineering #OpenScience
Closing the loop on #GenAI for #GPCR #SBDD www.nature.com/articles/s41... tinyurl.com/yeyhfr7j
Identification of nanomolar adenosine A2A receptor ligands using reinforcement learning and structure-based drug design - Nature Communications
Here the authors combine a deep generative model with structure-based drug design and prospectively validate functionally active, nanomolar, A2A adenosine receptor ligands and solve their crystal stru...
nature.com
Now in JCIM: pubs.acs.org/doi/full/10....
CACHE Challenge #2: Targeting the RNA Site of the SARS-CoV-2 Helicase Nsp13
AbstractA critical assessment of computational hit-finding experiments (CACHE) challenge was conducted to predict ligands for the SARS-CoV-2 Nsp13 helicase RNA binding site, a highly conserved COVID-1...
pubs.acs.org
Matthieu Schapira@mattschap.bsky.social · last yr.The CACHE #2 preprint is now online: bit.ly/3DyCmHN Active learning and fragment growing delivered confirmed hits. A citizen scientist using the Fold-it gaming interface designed the top compound! Kudos to Sasha and Madhushika @thesgc.bsky.social at the bench. @conscience-network.bsky.social
🚀 After two+ years of intense research, we’re thrilled to introduce Skala — a scalable deep learning density functional that hits chemical accuracy on atomization energies and matches hybrid-level accuracy on main group chemistry — all at the cost of semi-local DFT ⚛️🔥🧪🧬
CACHE4 results are out! All previously known CBLCB ligands shared the same scaffold. Congrats to Keunwan Park who successfully designed a chemically novel series, to the experimental team at @thesgc.bsky.social and thanks to @conscience-network.bsky.social for greasing the wheels! bit.ly/4mYNe3r
This week's cover of @rsc.org @chemicalscience.rsc.org AIMNet2: a neural network potential to meet your neutral, charged, organic, and elemental-organic needs. pubs.rsc.org/en/content/a... #compchem #chemsky
Excited to unveil Boltz-2, our new model capable not only of predicting structures but also binding affinities! Boltz-2 is the first AI model to approach the performance of FEP simulations while being more than 1000x faster! All open-sourced under MIT license! A thread… 🤗🚀
#compchem #machinelearning If you want to know more about #FeNNix-Bio1, the first foundation model able to perform accurate - long timescale- condensed phase molecular simulations of biological systems at quantum accuracy, join me in incoming live presentations: www.linkedin.com/feed/update/...
#fennix #machinelearning #gtc25 #cecam #watoc #vivatech #ai #docking #gpu… | Jean-Philip Piquemal
If you want to know more about #FeNNix-Bio1, the first #machinelearning foundation model able to perform accurate - long timescale- condensed phase molecular simulations of biological systems at quant...
linkedin.com
Our new preprint PharmacoForge: Pharmacophore Generation with Diffusion Models is out now! PharmacoForge quickly generates pharmacophores for a given protein pocket that identify key binding features and find useful compounds in a pharmacophore search. Check it out! 🧪 doi.org/10.26434/che...
The Open Molecules 2025 dataset is out! With >100M gold-standard ωB97M-V/def2-TZVPD calcs of biomolecules, electrolytes, metal complexes, and small molecules, OMol is by far the largest, most diverse, and highest quality molecular DFT dataset for training MLIPs ever made 1/N
#compchem New preprint: "A Foundation Model for Accurate Atomistic Simulations in Drug Design" FeNNix-Bio1, a foundation #machinelearning model for biosimulations doi.org/10.26434/che... #compchemsky #biosky Great work by T. Plé & the teams @lct-umr7616.bsky.social & @qubit-pharma.bsky.social
A Foundation Model for Accurate Atomistic Simulations in Drug Design
Neural network potentials now offer robust alternatives to electronic structure and empirical force fields computations for the on-the-fly production of the potential energy surfaces required in atomi...
doi.org
👋 🤖 Meet El Agente–an autonomous AI for performing computational chemistry, made by the Matter Lab @uoft.bsky.social. This #LLM-powered multi-agent system making computational chemistry more accessible will soon be available worldwide. Sign up 4 the launch: acceleration.utoronto.ca/news/meet-el...
@thesgc.bsky.social is generating large/open screening data and inviting data scientists to train their ML models via DREAM challenges: 1- train your model on DEL data 2- retrospectively predict 138 ASMS true positives 3- predict new hits. We will test them and publish together. bit.ly/3YXVKoT
First DREAM Target 2035 Drug Discovery Challenge
'First DREAM Target 2035 Drug Discovery Challenge' (Synapse ID: syn65660836) is a project on Synapse. Synapse is a platform for supporting scientific collaborations centered around shared biomedic...
bit.ly
"De novo prediction of protein structural dynamics" I'll be presenting an overview of the field tomorrow at a workshop. Link to a PDF copy of the presentation: delalamo.xyz/assets/post_...
delalamo.xyz
Encode protein structures as a series of discrete tokens, train a language model, and sample protein structural conformations given the sequence. arxiv.org/abs/2410.18403
AlphaFold is amazing but gives you static structures 🧊 In a fantastic teamwork, @mcagiada.bsky.social and @emilthomasen.bsky.social developed AF2χ to generate conformational ensembles representing side-chain dynamics using AF2 💃 Code: github.com/KULL-Centre/... Colab: github.com/matteo-cagia...
bioRxiv Biophysics@biorxiv-biophys.bsky.social · last yr.AF2χ: Predicting protein side-chain rotamer distributions with AlphaFold2 https://www.biorxiv.org/content/10.1101/2025.04.16.649219v1
New preprint: Finding Drug Candidate Hits With a Hundred Samples: Ultra-low Data Screening With Active Learning doi.org/10.26434/che... #compchem
The future of AI-drug discovery hinges on large, high-quality, standards-based datasets. No country or firm can build this alone. We need to construct datasets together in the open: www.science.org/doi/10.1126/... @tridentpct.bsky.social @conscience-network.bsky.social @mcgilluniversity.bsky.social
AI drug development’s data problem
The future of drug discovery may be artificial intelligence (AI), but its present is not. AI is in its infancy in the field. To help AI mature, developers need nonproprietary, open, large, high-qualit...
science.org
Run BioEmu in Colab - just click "Runtime → Run all"! Our notebook uses ColabFold to generate MSAs, BioEmu to predict trajectories, and Foldseek to cluster conformations. Thanks @jjimenezluna.bsky.social for the help! 🌐 colab.research.google.com/github/sokry... 📄 www.biorxiv.org/content/10.1...
Google Colab
colab.research.google.com
AlphaFold, the revolutionary, Nobel prize-winning tool for predicting protein structures, has a problem: it’s running low on data https://go.nature.com/3FJRyTd
AlphaFold is running out of data — so drug firms are building their own version
Thousands of 3D protein structures locked up in big-pharma vaults will be used to create a new AI tool that won’t be open to academics.
go.nature.com
🚀One week left to register for our Symposium on Open Drug Discovery! Join us in Montreal April 7-8 for an exciting program showcasing how open science and AI are driving drug discovery. Some sessions are already sold out, so register now! Register by April 2nd: conscience.ca/symposium2025
Conscience Symposium on Open Drug Discovery - Submission and registration are open! - Conscience
Registration is now open for the second annual Conscience Symposium on Open Drug Discovery! Join us at the Society for Arts and Technology in Montreal on April 7-8, 2025, for two days of insightful ta...
conscience.ca
This is a remarkable paper! A gigantic dataset of highly precise, highly accurate first-principles data. This builds on years of work on @fhi-aims.bsky.social - enabling dispersion-corrected hybrid DFT that covers a huge swath of chemical space. Congrats to the authors! doi.org/10.1038/s415...
The QCML dataset, Quantum chemistry reference data from 33.5M DFT and 14.7B semi-empirical calculations - Scientific Data
Scientific Data - The QCML dataset, Quantum chemistry reference data from 33.5M DFT and 14.7B semi-empirical calculations
urldefense.com
🚀 100 scientists. 31 countries. And we’re just getting started. #MAINFRAME is uniting global experts to drive AI-powered hit finding. ML models trained on real experimental data. Predictive tools tested in the lab. 🔗 Now is the time to join: aircheck.ai/mainframe
New Perspective on Community Benchmarking in Structure-Based Drug Design (SBDD)! #SBDD predictions need reliable benchmarks - diverse targets, high-quality affinity & structural data, and blinded validation. Let’s make it happen! 🔗 Read more: doi.org/10.1021/acs.... #DrugDiscovery #CompChem
The Need for Continuing Blinded Pose- and Activity Prediction Benchmarks
Computational tools for structure-based drug design (SBDD) are widely used in drug discovery and can provide valuable insights to advance projects in an efficient and cost-effective manner. However, d...
doi.org
The code & camera-ready version of our #ICLR2025 paper on "Multi-domain Distribution Learning for De Novo Drug Design" are now available 📚 Paper: openreview.net/forum?id=g3V... 💻 Code: github.com/LPDI-EPFL/Dr... (1/4)
