Hahah nice; but shouldn‘t it be „in silica“ (SiO2)? This would avoid the confusion ;-)

🤝 Lastly, huge thanks to the entire team: Ella Rajaonson, @chertianser.bsky.social, Mohammad Haddadnia, @shixuanleong.bsky.social, @aspuru.bsky.social, @agustinus.kristia.de, @valencekjell.com, @felix-s-k.bsky.social

💻 the Code is available openly at github.com/felix-s-k/cu...

Give it a try and let us know what you think!

🔎Lastly, we benchmark different Bayesian Optimization strategies and investigate how they can accelerate the identification of general reaction conditions. As an example, we find that the most efficient CurryBO strategy increases the identified generality 1.5x across a limited number of experiments.

🔎 Once formalized, we show that considering more substrates does indeed give more general reaction conditions that also work better for new substrates. While we augment common HTE datasets with more negative examples to be more realistic, this holds true for almost all datasets.

🔎 We tackle this challenge by formalizing the problem as an optimization problem over a curried function space — a concept from programming that reduces the number of input arguments of a function.

🔎 General reaction conditions are conditions that ‚work well‘ for multiple substrates; useful if you want to synthesize a library of compounds or need a substrate scope. While there is definitely a need for such conditions, noone systematically investigated how to obtain them in a data-driven way

neurips.cc/virtual/2024...

Could you add me as well please? 💻🧪

Could you add me as well please? Doing AI 4 Chemistry 🧪👨‍🔬