Javier Antorán

Super excited to announce @Angstrom_ai, our startup using Gen AI to build fast and experimentally accurate simulations of molecular interactions together with @SilkyDogfish, @jmhernandez233 and Gabor Csanyi. We are backed by @ycombinator! See below for a simulation demo (1/3)🧵.

I'll be at NeurIPS next week, presenting our work "A Generative Model of Symmetry Transformations." In it, we propose a symmetry-aware generative model that discovers which (approximate) symmetries are present in a dataset, and can be leveraged to improve data efficiency. 🧵⬇️

It was such a pleasure co-organizing @aabi_org this year! We had a great program---with fantastic posters, speakers, and panelists---and talked about topics ranging from approximate inference and smart priors to AI safety and LLMs! 🚀Check out the first-ever AABI proceedings!🚀

Proceedings of the 6th Symposium on Advances in Approximate Bayesian Inference, held in Vienna and co-located with #ICML2024, is now available at proceedings.mlr.press/v253/ #AABI #Bayes

Excited to kick off the 6th Symposium on Advances in Approximate Bayesian Inference! We will be livestreaming the event here: bit.ly/aabi2024 🚀 Check out the schedule and our many great speakers: approximateinference.org/schedule/index… #ICML2024 @aabi_org

If you can't be in Vienna today to join us for AABI 2024, you can watch our live stream!

Really excited that AABI is sponsored by Basis this year. See you on Sunday!

🚀 @Angstrom_ai launched! Replace wetlab experiments with Gen AI molecular simulations "Accelerating molecular simulation using generative AI" ⚡️ 🌐 angstrom-ai.com 🧪 Ångström AI builds fast and experimentally accurate generative AI-based simulations of molecular interactions for pharma and biotech companies 💊 💡 These simulations can determine whether a drug is going to bind to a protein or how quickly a drug will act once it is consumed by a patient. The simulations obey the laws of physics, avoiding the hallucinations seen in other GenAI technologies. ✅ Compute free energy differences, binding conformations & hydration sites, with ab initio accuracy, orders of magnitude faster than traditional simulations. 📊 Quantum mechanically accurate models of physics + generative AI that allows them to run these models quickly: Their models generate states consistent with physics, but the transitions between states are non-physical and significantly faster. Diffusion models accelerate MACE simulations, making them computationally affordable. 🔬 To start off they are focusing on providing experimentally accurate estimates of solubility, which is important for determining the bioavailability of drugs. Poor solubility causes many computationally designed drugs to fail experimental validation! 🧠 They have successfully sped up Supercool water & Hydrating methane: These are the first-ever GenAI-accelerated, physically accurate molecular dynamics simulations incorporating the interaction of many molecules. Ångström AI has published the first molecule water solubility results with accuracy within the error range of wet lab experiments. 🗓 Interested in a demo? Reach out and book a call! 👥 Have friends who work in pharma or biotech interested in computational methods? Make the intro! Congrats on the launch @JaviAC7, @SilkyDogfish, @jmhernandez233, & Gabor Csanyi!! tryfondo.com/blog/angstroma…

Registration for #AABI2024 in 📷Vienna📷 (co-located with ICML) is now open at aabi2024.eventbrite.com !! Spots are free but limited, so get your tickets quick!⏩

Going to ICML is ~2 weeks? In Vienna already on Sunday? Why not hang out with probabilistic inference folks? Chat about the future of prob. modeling in the foundation model era, decision-making & planning under uncertainty, and other stuff! Registration is free, but limited 👇

Super stoked to be building accelerated molecular simulations at @Angstrom_ai with @JaviAC7 @jmhernandez233 and Gabor Csanyi! We combine experimentally accurate force fields (MACE) with accelerated sampling using generative models. To start off we are focusing on providing experimentally accurate estimates of solubility, which is important for determining the bioavailability of drugs. Poor solubility causes many ML designed drugs to fail experimental validation!

YC S24's @Angstrom_ai builds fast and experimentally accurate Gen AI simulations of molecular interactions, replacing wet lab experiments in the drug development pipeline. ycombinator.com/launches/LGp-n… Congraats on the launch, @JaviAC7, @SilkyDogfish, and @jmhernandez233!

In hydration simulations, the slowest modes are given by the interaction between the solute and its first water shell. Our models draw independent and unbiased samples from this region, greatly accelerating the convergence of free energy estimations. (3/3)

In de novo drug design, proposals often fail during experimental validation due to poor solubility. That is why we are starting by using our tech to accelerate the computation of  hydration free energies! (2/3)

Super excited to announce @Angstrom_ai, our startup using Gen AI to build fast and experimentally accurate simulations of molecular interactions together with @SilkyDogfish, @jmhernandez233 and Gabor Csanyi. We are backed by @ycombinator! See below for a simulation demo (1/3)🧵.

@eeevgen @aabi_org We have been trying to recover it. We have not yet succeeded.

@JaviAC7 Why @aabi_org Twitter is dead? If it is dead why is it on current year site?

TLDR: @JihaoAndreasLin continues to make GPs Better, Faster, Stronger. GP-based LLM coming soon. 🤖🤖 We review and analyze best-practices for GP hyper parameter learning which are not given attention in the literature, but when combined, yield orders of magnitude speedups!

@JamesAllingham @PetarV_93 @predict_addict He hid all the posts where I attempted to respond :(

@PetarV_93 Thanks for calling this out Petar! It’s a shame that @predict_addict has no interest in actually engaging…

Disagree all you want with Bayesian learning, but there are _far_ better ways to do it than screenshotting a PhD graduate's thesis and shaming an entire university. I didn't study Bayesian methods at Cambridge but I interacted with many Bayesians and learnt _so much_ from them.

@PetarV_93 @amirvaxman_dgp @mmbronstein Bayesian inference is all about averaging dissenting hypotheses, and this guy isn't having any of that! @PetarV_93 @mmbronstein @roydanroy @amirvaxman_dgp

@amirvaxman_dgp I suspect they may be blocking everyone who likes any post in this chain. I see you liked @mmbronstein's.

@BernabeLou @PetarV_93 Note that my thesis was privately funded! You can see the funding source in the acknowledgements.

@PetarV_93 I also find this kind of shaming rather improper, even though I can agree with the stated objective of saving taxpayer money.