Data-driven Insights on the Impact of Functionalization on Metal-Organic Framework (MOF) Free Energies

Functionalization is poised to play a prominent role in MOF development as it could become the to-go strategy to bestow extant MOF with new properties, and to control MOF pore shape and size by modulating polymorph selection. Thus, to speed up MOF development through computational work, a better (predictive) understanding on how functionalization impacts MOF synthesizability is needed. Here we use a data-driven approach where molecular dynamics simulations on 5,000+ MOFs are used to shed light on how functionalization affects MOF free energy, as the latter has been largely tied to MOF synthesizability and polymorph selection. More consistently in MOFs with higher void fractions, we find that functionalization generally reduces free energy, with entropy contributing significantly to this thermodynamic stabilization. Although with some functionalizations (-CF3, -F, -Br, -SH, -OH) the role of entropy is more apparent than with others (-CN, -CH3, -NO2, -NH3). Through uneven stabilization of polymorphs, we also find functionalization (more often with -Br, -CN and -CF3) as capable of altering polymorph (topology) selection relative to original non-functionalized polymorphic families. However, no switch in polymorph stability ever occurred when the original (unfunctionalized) polymorphs were separated by more than 1.42 kJ/mol per MOF atom. We show that machine learning can predict functionalization-induced free energy change of a parent MOF with a mean absolute error of 0.16 kJ/mol per atom, using only physical properties of the parent MOF and the functional group as input. The ML-based SHAP analysis agrees with human analysis on the functionalization molecular mass and the hydrogen fraction of the parent MOF being among the factors that influence change in free energy the most. Finally, we present a publicly accessible dynamic interface to visualize and navigate the free energy data , thereby encouraging the research community to engage with and utilize the data to help uncover new insights.