DMSE PhD candidate Soojung Yang was among 13 newly selected Takeda Fellows for the 2023-24 academic year. Yang's research applies methods in geometric deep learning and generative modeling to better understand protein dynamics. buff.ly/47iBkJg

It was wonderful to meet all 300+ of you at #MoML2023! 😃 So exciting to see the budding potential of early-stage molecular ML research & predict what the rapidly approaching future of molecular ML could look like in just a few years. See you all next year! 👋

Thrilled to share my recently published work describing another curious case of inter-domain horizontal gene transfer (HGT), this time involving nematodes! HGT is just 😍🤯🔥 Many thanks to Matt Daugherty and a wonderful collaboration with Dengke Ma! embopress.org/doi/full/10.15…

I'll be at #NeurIPS2023 next week, presenting an Oral on our recent work introducing a classic psychophysics method to the study of deep net robustness! If you're also interested in NeuroAI, robustness and alignment, DM me and we can find a time to chat! neurips.cc/virtual/2023/o…

I'm happy to share that our paper "Autonomous, multiproperty-driven molecular discovery: From predictions to measurements and back" was published yesterday in Science Magazine: science.org/doi/10.1126/sc…

Professors Bilge Yildiz and Rafael Gomez-Bombarelli are using AI to design new compounds or alloys whose surfaces can be used as catalysts in chemical reactions. Their research was published in Nature Computational Science. buff.ly/48EyfUt Nature Computational Science

Zeolites, porous materials with diverse applications, often require intricate synthesis processes. Elton Pan Olivetti Group @ MIT and colleagues MIT ChemE Dept and ITQ (UPV-CSIC) developed an open-source dataset of 23,000+ synthesis routes. Read more in ACS Central Science: buff.ly/43bGWUD

Check out our recent piece talking about the 'execution gap' between genAI modeling and real-world applications in chemistry and materials science. We also provide thoughts on paths forward to balance development with risk and mitigation strategies.

📢 New preprint! Our latest work enables "alchemy" - ∂[energy]/∂[element] in ML potentials like MACE. We model solid solutions and conduct alchemical free energy simulations. arxiv.org/abs/2404.10746 RGB Lab @ MIT #compchem #machinelearning

Excited to share that I will be interning with Samsung Display over the Summer, where I will be working on generative models and atomistic simulations to design molecules for OLEDs! If you are in the San Jose/Bay area and would like to hang out over the Summer, let me know!

Excited to start my summer internship this week at Lawrence Livermore National Laboratory LLNL Computing where I'll be working on studying robustness of machine and human vision! If you are also in the Bay Area over the summer and/or have similar interests, let's meet up!

The same reason we're seeing less of RL and proxy rewards in AI for science. We're able to somewhat deal with this by 'restricting' the generator through physical constraints, or 'expanding' the proxy reward through active learning. Surprised it even works on unconstrained LLMs!