Large language models (LLMs) have gained widespread interest due to their ability to process human language and perform tasks on which they have not been explicitly trained. This is relevant for the chemical sciences, which face the problem of small and diverse datasets that are frequently in the form of text. LLMs have shown promise in addressing these issues and are increasingly being harnessed to predict chemical properties, optimize reactions, and even design and conduct experiments autonomously. However, we still have only a very limited systematic understanding of the chemical reasoning capabilities of LLMs, which would be required to improve models and mitigate potential harms. Here, we introduce "ChemBench," an automated framework designed to rigorously evaluate the chemical knowledge and reasoning abilities of state-of-the-art LLMs against the expertise of human chemists. We curated more than 7,000 question-answer pairs for a wide array of subfields of the chemical sciences, evaluated leading open and closed-source LLMs, and found that the best models outperformed the best human chemists in our study on average. The models, however, struggle with some chemical reasoning tasks that are easy for human experts and provide overconfident, misleading predictions, such as about chemicals' safety profiles. These findings underscore the dual reality that, although LLMs demonstrate remarkable proficiency in chemical tasks, further research is critical to enhancing their safety and utility in chemical sciences. Our findings also indicate a need for adaptations to chemistry curricula and highlight the importance of continuing to develop evaluation frameworks to improve safe and useful LLMs.
Large Language Models (LLMs) create exciting possibilities for powerful language processing tools to accelerate research in materials science. While LLMs have great potential to accelerate materials understanding and discovery, they currently fall short in being practical materials science tools. In this position paper, we show relevant failure cases of LLMs in materials science that reveal current limitations of LLMs related to comprehending and reasoning over complex, interconnected materials science knowledge. Given those shortcomings, we outline a framework for developing Materials Science LLMs (MatSci-LLMs) that are grounded in materials science knowledge and hypothesis generation followed by hypothesis testing. The path to attaining performant MatSci-LLMs rests in large part on building high-quality, multi-modal datasets sourced from scientific literature where various information extraction challenges persist. As such, we describe key materials science information extraction challenges which need to be overcome in order to build large-scale, multi-modal datasets that capture valuable materials science knowledge. Finally, we outline a roadmap for applying future MatSci-LLMs for real-world materials discovery via: 1. Automated Knowledge Base Generation; 2. Automated In-Silico Material Design; and 3. MatSci-LLM Integrated Self-Driving Materials Laboratories.
Recent advances in computational modelling of atomic systems, spanning molecules, proteins, and materials, represent them as geometric graphs with atoms embedded as nodes in 3D Euclidean space. In these graphs, the geometric attributes transform according to the inherent physical symmetries of 3D atomic systems, including rotations and translations in Euclidean space, as well as node permutations. In recent years, Geometric Graph Neural Networks have emerged as the preferred machine learning architecture powering applications ranging from protein structure prediction to molecular simulations and material generation. Their specificity lies in the inductive biases they leverage -- such as physical symmetries and chemical properties -- to learn informative representations of these geometric graphs. In this opinionated paper, we provide a comprehensive and self-contained overview of the field of Geometric GNNs for 3D atomic systems. We cover fundamental background material and introduce a pedagogical taxonomy of Geometric GNN architectures:(1) invariant networks, (2) equivariant networks in Cartesian basis, (3) equivariant networks in spherical basis, and (4) unconstrained networks. Additionally, we outline key datasets and application areas and suggest future research directions. The objective of this work is to present a structured perspective on the field, making it accessible to newcomers and aiding practitioners in gaining an intuition for its mathematical abstractions.
Sampling diverse, thermodynamically feasible molecular conformations plays a crucial role in predicting properties of a molecule. In this paper we propose to use GFlowNet for sampling conformations of small molecules from the Boltzmann distribution, as determined by the molecule's energy. The proposed approach can be used in combination with energy estimation methods of different fidelity and discovers a diverse set of low-energy conformations for highly flexible drug-like molecules. We demonstrate that GFlowNet can reproduce molecular potential energy surfaces by sampling proportionally to the Boltzmann distribution.
Structure determination is necessary to identify unknown organic molecules, such as those in natural products, forensic samples, the interstellar medium, and laboratory syntheses. Rotational spectroscopy enables structure determination by providing accurate 3D information about small organic molecules via their moments of inertia. Using these moments, Kraitchman analysis determines isotopic substitution coordinates, which are the unsigned $|x|,|y|,|z|$ coordinates of all atoms with natural isotopic abundance, including carbon, nitrogen, and oxygen. While unsigned substitution coordinates can verify guesses of structures, the missing $+/-$ signs make it challenging to determine the actual structure from the substitution coordinates alone. To tackle this inverse problem, we develop KREED (Kraitchman REflection-Equivariant Diffusion), a generative diffusion model that infers a molecule's complete 3D structure from its molecular formula, moments of inertia, and unsigned substitution coordinates of heavy atoms. KREED's top-1 predictions identify the correct 3D structure with >98% accuracy on the QM9 and GEOM datasets when provided with substitution coordinates of all heavy atoms with natural isotopic abundance. When substitution coordinates are restricted to only a subset of carbons, accuracy is retained at 91% on QM9 and 32% on GEOM. On a test set of experimentally measured substitution coordinates gathered from the literature, KREED predicts the correct all-atom 3D structure in 25 of 33 cases, demonstrating experimental applicability for context-free 3D structure determination with rotational spectroscopy.
We propose an instruction-based process for trustworthy data curation in materials science (MatSci-Instruct), which we then apply to finetune a LLaMa-based language model targeted for materials science (HoneyBee). MatSci-Instruct helps alleviate the scarcity of relevant, high-quality materials science textual data available in the open literature, and HoneyBee is the first billion-parameter language model specialized to materials science. In MatSci-Instruct we improve the trustworthiness of generated data by prompting multiple commercially available large language models for generation with an Instructor module (e.g. Chat-GPT) and verification from an independent Verifier module (e.g. Claude). Using MatSci-Instruct, we construct a dataset of multiple tasks and measure the quality of our dataset along multiple dimensions, including accuracy against known facts, relevance to materials science, as well as completeness and reasonableness of the data. Moreover, we iteratively generate more targeted instructions and instruction-data in a finetuning-evaluation-feedback loop leading to progressively better performance for our finetuned HoneyBee models. Our evaluation on the MatSci-NLP benchmark shows HoneyBee's outperformance of existing language models on materials science tasks and iterative improvement in successive stages of instruction-data refinement. We study the quality of HoneyBee's language modeling through automatic evaluation and analyze case studies to further understand the model's capabilities and limitations. Our code and relevant datasets are publicly available at \url{https://github.com/BangLab-UdeM-Mila/NLP4MatSci-HoneyBee}.
The use of machine learning for material property prediction and discovery has traditionally centered on graph neural networks that incorporate the geometric configuration of all atoms. However, in practice not all this information may be readily available, e.g.~when evaluating the potentially unknown binding of adsorbates to catalyst. In this paper, we investigate whether it is possible to predict a system's relaxed energy in the OC20 dataset while ignoring the relative position of the adsorbate with respect to the electro-catalyst. We consider SchNet, DimeNet++ and FAENet as base architectures and measure the impact of four modifications on model performance: removing edges in the input graph, pooling independent representations, not sharing the backbone weights and using an attention mechanism to propagate non-geometric relative information. We find that while removing binding site information impairs accuracy as expected, modified models are able to predict relaxed energies with remarkably decent MAE. Our work suggests future research directions in accelerated materials discovery where information on reactant configurations can be reduced or altogether omitted.
Reinforcement learning (RL) over text representations can be effective for finding high-value policies that can search over graphs. However, RL requires careful structuring of the search space and algorithm design to be effective in this challenge. Through extensive experiments, we explore how different design choices for text grammar and algorithmic choices for training can affect an RL policy's ability to generate molecules with desired properties. We arrive at a new RL-based molecular design algorithm (ChemRLformer) and perform a thorough analysis using 25 molecule design tasks, including computationally complex protein docking simulations. From this analysis, we discover unique insights in this problem space and show that ChemRLformer achieves state-of-the-art performance while being more straightforward than prior work by demystifying which design choices are actually helpful for text-based molecule design.
Equivariant graph neural networks force fields (EGraFFs) have shown great promise in modelling complex interactions in atomic systems by exploiting the graphs' inherent symmetries. Recent works have led to a surge in the development of novel architectures that incorporate equivariance-based inductive biases alongside architectural innovations like graph transformers and message passing to model atomic interactions. However, thorough evaluations of these deploying EGraFFs for the downstream task of real-world atomistic simulations, is lacking. To this end, here we perform a systematic benchmarking of 6 EGraFF algorithms (NequIP, Allegro, BOTNet, MACE, Equiformer, TorchMDNet), with the aim of understanding their capabilities and limitations for realistic atomistic simulations. In addition to our thorough evaluation and analysis on eight existing datasets based on the benchmarking literature, we release two new benchmark datasets, propose four new metrics, and three new challenging tasks. The new datasets and tasks evaluate the performance of EGraFF to out-of-distribution data, in terms of different crystal structures, temperatures, and new molecules. Interestingly, evaluation of the EGraFF models based on dynamic simulations reveals that having a lower error on energy or force does not guarantee stable or reliable simulation or faithful replication of the atomic structures. Moreover, we find that no model clearly outperforms other models on all datasets and tasks. Importantly, we show that the performance of all the models on out-of-distribution datasets is unreliable, pointing to the need for the development of a foundation model for force fields that can be used in real-world simulations. In summary, this work establishes a rigorous framework for evaluating machine learning force fields in the context of atomic simulations and points to open research challenges within this domain.
We propose MatSci ML, a novel benchmark for modeling MATerials SCIence using Machine Learning (MatSci ML) methods focused on solid-state materials with periodic crystal structures. Applying machine learning methods to solid-state materials is a nascent field with substantial fragmentation largely driven by the great variety of datasets used to develop machine learning models. This fragmentation makes comparing the performance and generalizability of different methods difficult, thereby hindering overall research progress in the field. Building on top of open-source datasets, including large-scale datasets like the OpenCatalyst, OQMD, NOMAD, the Carolina Materials Database, and Materials Project, the MatSci ML benchmark provides a diverse set of materials systems and properties data for model training and evaluation, including simulated energies, atomic forces, material bandgaps, as well as classification data for crystal symmetries via space groups. The diversity of properties in MatSci ML makes the implementation and evaluation of multi-task learning algorithms for solid-state materials possible, while the diversity of datasets facilitates the development of new, more generalized algorithms and methods across multiple datasets. In the multi-dataset learning setting, MatSci ML enables researchers to combine observations from multiple datasets to perform joint prediction of common properties, such as energy and forces. Using MatSci ML, we evaluate the performance of different graph neural networks and equivariant point cloud networks on several benchmark tasks spanning single task, multitask, and multi-data learning scenarios. Our open-source code is available at https://github.com/IntelLabs/matsciml.