Robust actin filament modeling in low-resolution cryo-ET images using advanced deep learning and algorithmic techniques

The National Institute of General Medical Sciences awarded Georgia Southern University Research & Service Foundation Inc. $142,000 on July 16, 2026, under the Biomedical Research and Research Training program (CFDA 93.859) to develop computational tools for analyzing actin filament architecture in cryo-electron tomography (cryo-ET) images.

The project addresses a critical bottleneck in structural biology: current methods for extracting quantitative morphological data from cryo-ET images are severely hampered by low signal-to-noise ratios, missing wedge artifacts, and the absence of robust automated computational tools for modeling complex actin networks. The research will develop and validate an integrated computational framework combining filament-specific image enhancement, advanced deep learning segmentation, and geometry-aware reconstruction. Three specific aims structure the work: developing a local dynamic programming algorithm to enhance contrast of filamentous structures by leveraging their linear geometry; creating a dedicated deep learning segmentation tool pre-trained on large-scale simulated datasets with realistic noise profiles and fine-tuned on experimental data; and applying geometry-aware reconstruction approaches to improve accuracy.

The award supports research into dysregulation of the actin cytoskeleton, which drives pathogenesis in cancer metastasis, cardiovascular dysfunction, and neurodegenerative conditions. Work is performed in Georgia. The period of performance runs from July 16, 2026, to April 30, 2030. The assistance type is a Project Grant.

Abstract

Dysregulation of the actin cytoskeleton drives the pathogenesis of chronic diseases, including cancer metastasis, cardiovascular dysfunction, and neurodegenerative conditions, which collectively contribute to over 30 million deaths annually. The ability to quantitatively characterize actin filament architecture within its native cellular context is essential for understanding the structural basis of this dysregulation. Cryo-electron tomography (cryo-ET) has emerged as a uniquely powerful technique for visualizing macro- molecular complexes like actin filaments within their native cellular context; however, extracting accurate, quantitative morphological data is severely hampered by technical challenges. These include intrinsically low signal-to-noise ratios exacerbated by the missing wedge artifact, and a critical lack of robust, auto- mated computational tools modeling the complex, densely packed actin networks. This project proposes to address these limitations by developing and validating a novel, integrated computational framework specifically designed for the analysis of actin filaments from cryo-ET data. Our central hypothesis is that a tailored approach combining filament-specific image enhancement, advanced deep learning, and geometry-aware reconstruction will overcome bottlenecks and enable unprecedented quantitative assessment of actin morphology in native states. To test this hypothesis, we will pursue three synergistic specific aims: (1) we will develop a local dynamic programming (DP) algorithm to enhance the contrast of filamentous structures by leveraging their linear geometry to amplify coherent signals against noise. (2) we will create a dedicated deep learning segmentation tool, pre-trained on large- scale simulated datasets with realistic noise profiles and fine-tuned on experimental data, to achieve accurate and generalizable identification of actin filaments across cellular contexts. (3) we will implement a reconstruction pipeline that incorporates filament-specific clustering and directional analysis to convert segmented volumes into discrete, biologically plausible 3D models. The successful completion of these aims will deliver a comprehensive, open-source computational framework for automated and reproducible analysis of actin network architecture, helping to elucidate the connection between structural alterations and disease-specific outcomes. The expected outcomes include mechanistic insights into actin-driven disease processes, laying the groundwork for future diag- nostic and therapeutic innovations. Furthermore, this interdisciplinary project aligns with the NIH support for research excellence (sure) mission by providing hands-on research training for undergraduate and graduate students, thereby strengthening biomedical research capacity at georgia southern university.