Publications

Molecular Foundation Models are emerging as powerful tools for accelerating molecular design, material science, and cheminformatics, leveraging transformer architectures to speed up the discovery of new materials and drugs while reducing the computational cost of traditional ab initio methods. However, current models are constrained by closed-vocabulary tokenizers that fail to capture the full diversity of molecular structures. In this work, we systematically evaluate thirteen chemistry-specific tokenizers for their coverage of the SMILES language, uncovering substantial gaps. Using N-gram language models, we accessed the impact of tokenizer choice on model performance and quantified the information loss of unknown tokens. We introduce two new tokenizers, \emph{smirk} and \emph{smirk-gpe}, which can represent the entirety of the OpenSMILES specification while avoiding the pitfalls of existing tokenizers. Our work highlights the importance of open-vocabulary modeling for molecular foundation models and the need for chemically diverse benchmarks for cheminformatics.

Enabling widespread electric vehicle (EV) adoption requires a substantial build-out of charging infrastructure in the coming decade. We formulate the charging infrastructure needs as a scaling analysis problem and use it to estimate the EV infrastructure needs of the USA at a county-level resolution. We find that gasoline and EV charging stations scale sub-linearly with their respective vehicle registrations, recovering the sub-linear scaling typical of infrastructure. Surprisingly, we find that EV charging stations scale super-linearly with population size within counties, deviating from the sub-linear scaling of gasoline stations. We discuss how this demonstrates the infancy of both EVs and EV infrastructure while providing a framework for estimating future EV infrastructure demands. By considering the power delivery of existing gasoline stations, and appropriate EV efficiencies, we estimate the EV infrastructure gap at the county level, providing a road map for future EV infrastructure expansion.

Electric vertical takeoff and landing aircraft have a unique duty cycle characterized by high discharge currents at the beginning and end of the mission (corresponding to takeoff and landing of the aircraft) and a moderate power requirement between them with no rest periods during the mission. Here, we generated a dataset of battery duty profiles for an electric vertical takeoff and landing aircraft using a cell typical for that application. The dataset features 22 cells, comprising a total of 21,392 charge and discharge cycles. 3 of the cells use the baseline cycle while each of the other cells vary either charge current, discharge power, discharge duration, ambient cooling conditions, or end of charge voltage. While it was designed to mimic the expected duty cycle of an electric aircraft, this dataset is relevant for training machine learning models on battery life, fitting physical or empirical models for battery performance and/or degradation, and countless other applications.