From ions to cells: integrating machine learning and cryoem to resolve macromolecular complexity

The National Institute of General Medical Sciences awarded the University of Texas Health Science Center at Houston $423,576 on July 1, 2026, under the Biomedical Research and Research Training program (CFDA 93.859) to develop machine learning-driven computational tools for interpreting electron cryo-microscopy and electron cryo-tomography data of complex biological assemblies.

The project addresses significant gaps in current cryo-EM modeling software by building a unified framework to model biological systems containing non-protein components (lipids, ions, and nucleic acids), structural heterogeneity, and large-scale ultrastructural changes. Development will be driven by four exemplar systems: modeling ligand, lipid, and ion interactions in membrane proteins; identifying and modeling nucleic acids in topoisomerases; resolving structural heterogeneity among prion conformational states; and enabling structural discovery in cellular cryo-ET datasets of an immune synapse. These research areas have translational relevance to cancer, neurological disorders, cardiovascular disease, and chimeric antigen receptor T cell engineering.

The award carries a performance period through April 30, 2031, with place of performance in Houston, Texas. The recipient operates through its Sponsored Projects Administration Division as part of the University of Texas System.

Abstract

Electron cryo-microscopy (cryoem) and electron cryo-tomography (cryoet) have become indispensable tools for determining macromolecular structure and function. However, significant challenges remain in interpreting and modeling complex biological assemblies, particularly in systems that are structurally heterogeneous, contain non-protein components (lipids, ions, and nucleic acids), or undergo large-scale ultrastructural changes. This project aims to develop a unified, machine learning-driven modeling framework to address these limitations, enabling more accurate interpretation of cryoem and cryoet data and revealing new mechanistic insights into functional mechanisms that govern cellular processes and disease. To build computational tools that support robust modeling of complex, biologically relevant systems, development will be driven by four exemplar systems where current cryoem modeling software falls short: (1) modeling the interactions of ligands, lipids, and ions in membrane proteins; (2) identifying and modeling nucleic acids in topoisomerases; (3) resolving structural heterogeneity among prion conformational states; and (4) enabling structural discovery in cellular cryoet datasets of an immune synapse. These projects represent the type of biological complexity and resolution-specific detail that must be mined from contemporary and future cryoem/cryoet datasets, making them ideal testbeds for developing generalizable modeling frameworks. Moreover, each project has strong translational relevance, with clear implications for human disease and therapeutic development. Accurate modeling of membrane protein interactions with lipids, ions, and small molecules will deepen our understanding of diseases such as cancer, neurological disorders, and cardiovascular disease. In situ modeling of the immune synapse formed by chimeric antigen receptor (car) t cells and tumor cells will inform the design of next-generation car t-cell therapies and address key challenges in treating solid tumors. Structural characterization of prion polymorphism will help define the molecular basis of β-amyloid fibril formation, a hallmark of multiple neurodegenerative disorders. Finally, elucidating the components that govern topoisomerase recognition and activity will shed light on critical mechanisms of DNA topology regulation and genome maintenance, with relevance to cancer and the development of topoisomerase-targeted therapeutics. By developing and disseminating these next-generation computational tools, we will equip researchers with the resources needed to interpret increasingly complex biological data. Building on representative systems, our modeling framework will generalize across diverse types of macromolecular assemblies, enabling accurate interpretation of features that have historically been inaccessible. This will accelerate structural discovery in complex systems and ensure that resulting models are both experimentally grounded and biologically meaningful. Ultimately, this work will support new strategies in precision medicine by providing structural insights into disease-relevant molecular pathways and opening new avenues for therapeutic development.