Break

Kunhuan Liu is interested in the accurate prediction and optimization of structural and functional properties of new materials, guided by theory and computational techniques. His work focuses on the development of in silico structural databases and machine learning-driven high-throughput screening workflows. Integrating cheminformatics, molecular simulation, and data-driven modeling, his research has systematically investigated the fundamental design parameters necessary for metal-organic frameworks to achieve ultrahigh volumetric capacities for hydrogen storage.

During his internship at Poseida Therapeutics Inc., he assisted the nanoparticle delivery team by incorporating computational screening into the experimental workflow. This included the use of quantitative structure–property relationship (QSPR) models and computationally derived descriptors for statistical learning to design and optimize nanoparticle formulations. With a strong passion for numerical modeling and structure optimization through simulation, Kunhuan is actively seeking full-time employment starting in January 2025.

Abstract:

Hydrogen is considered a crucial clean energy vector to mitigate climate change, but due to the low volumetric energy density of gaseous hydrogen, it is difficult to store hydrogen for many practical applications. Cryogenic sorption-based methods, particularly using metal–organic frameworks (MOFs), have been recognized as viable solutions to enhance the deliverable capacity of stored hydrogen. The first part of my work focused on two main challenges: identifying materials that surpass the highest recorded performance for volumetric deliverable capacity, currently held by MOF-5, the benchmark material for the past two decades, and clarifying how optimal structural properties translate into actionable design parameters, such as organic linkers, metal clusters, and underlying topologies, which define how the chemical building blocks are connected.

To address these challenges, I adopted a data-driven approach that integrated molecular simulations, high-throughput screening, and machine learning. To systematically explore the effect of the MOF topology, I constructed in silico 105,764 MOF structures using 534 topologies and performed high-throughput screening of their hydrogen deliverable capacities using grand canonical Monte Carlo (GCMC) simulations and surrogate machine learning models. Analyzing over 100,000 MOFs, we explored the less-known effect of the underlying topologies of MOFs on hydrogen deliverable capacity. I subsequently identified the key mathematical descriptors of the topologies that explain the correlation between topologies, structural properties, and volumetric deliverable capacity. Our findings uncovered design principles to target specific MOF topologies for ultrahigh capacity hydrogen storage.

On the material discovery front, the machine learning model achieved state-of-the-art accuracy in predicting deliverable capacities. This model successfully identified all the top performing MOF materials in the dataset with an excellent agreement to GCMC simulations. Subsequently, I identified the most promising candidates with synthetic feasibility and simulated capacities surpassing the record performance.

In the second part of my work, I studied rht-MOFs that are based on supermolecular building blocks (SBBs). I showed that computational techniques can be used to explore a variety of SBBs, which are otherwise experimentally expensive to study. Our team synthesized the best performing rht-MOF and validated our simulation results. Furthermore, we analyzed how the varying cavity size and chemical composition of the SBBs affect the adsorbed hydrogen per unit surface area, thereby elucidating the structure–property relationships.