SMART (Shuimu Automated Reconstruction Technology): An Integrated Computational Platform for Streamlining Cryo-Electron Microscopy Workflow

Summary

This protocol demonstrates a step-by-step cryo-electron microscopy workflow using an integrated software platform (SMART) that connects data acquisition, three-dimensional reconstruction, and atomic model building in a single browser-based interface. As a representative example, the human TRPML1 structure is determined at 2.38 Å resolution.

Abstract

Cryo-electron microscopy (cryo-EM) has become a central tool in structural biology, yet current workflows typically require multiple specialized software packages for data acquisition, three-dimensional (3D) reconstruction, and atomic model building, leading to fragmented pipelines and frequent manual intervention. Here, we present SMART (Shuimu Automated Reconstruction Technology), an integrated software platform that combines three modules—DataSmart for automated data collection, CryoSmart for image processing and 3D reconstruction, and ModelSmart for deep-learning-based atomic model building—within a unified browser-accessible interface. This protocol provides step-by-step instructions for operating all three modules. The workflow encompasses automated specimen navigation and data acquisition, motion correction, contrast transfer function (CTF) estimation, particle picking, two-dimensional (2D) classification, 3D refinement, map enhancement, and atomic model generation. As a representative example, we applied the complete workflow to determine the structure of human TRPML1, a Ca²⁺-permeable lysosomal cation channel, achieving a global resolution of 2.38 Å by gold-standard Fourier shell correlation (FSC) at the 0.143 criterion. All three modules (DataSmart, CryoSmart, and ModelSmart) were applied sequentially to the same TRPML1 dataset within this study. The protocol is designed to be accessible to users with varying levels of cryo-EM experience.

Introduction

Cryo-Electron Microscopy: Technological Evolution and Biological Impact. The determination of three-dimensional protein structure remains a central challenge in molecular biology, as biological function is intricately linked to the spatial arrangement of amino acid residues. Elucidating these structural configurations is imperative for deciphering enzymatic mechanisms, ligand interactions, and cellular signaling pathways at atomic resolution. This imperative has driven transformative developments in cryo-EM, which has emerged as a widely adopted structural biology technique^1,2.

Cryo-EM single-particle analysis (SPA) serves as a powerful approach for obtaining biological insights across diverse sample types by resolving high-resolution structures of isolated complexes³. These include viruses, membrane proteins, helical assemblies, and other dynamic and heterogeneous macromolecular complexes, spanning a broad range of molecular sizes—from approximately 50 kDa to tens of megadaltons.

Cryo-EM SPA has emerged as a cornerstone technique for resolving the three-dimensional (3D) architectures of biological macromolecules, encompassing three interdependent computational stages: data acquisition, 3D reconstruction, and atomic model building (Figure 1).

During data acquisition, vitrified samples are imaged under low-dose conditions to generate micrograph movies, capturing thousands of particle projections in near-native states. These raw datasets are subsequently processed through iterative alignment and classification algorithms to reconstruct volumetric density maps. Finally, atomic models are built and refined to fit the density, elucidating molecular interactions at near-atomic resolution.

The analysis above demonstrates that output of cryo-EM merely represents the starting point of the structural pipeline. Transforming these raw data into interpretable atomic models demands a cascade of specialized computational software tools. Thus, software serves as the critical bridge transforming raw data into atomic models.

Fragmented cryo-EM Software Ecosystem: Current Tools and Limitations. Modern cryo-EM workflows rely on a patchwork of specialized software tools, each addressing discrete stages:

Data Acquisition. While numerous cryo-EM data acquisition software packages are publicly available—including Leginon^4,5, SerialEM⁶, UCSFImage4⁷, TFS EPU^8,9, Gatan Latitude, JEOL JADAS¹⁰, and Auto-EMation¹¹—these tools predominantly output raw movie data with limited post-acquisition processing capabilities. Critical workflow steps, such as parameter setting, template customization, still demand labor-intensive manual operations, often requiring too many parameters per dataset. This reliance on expert intervention introduces reproducibility challenges and workflow inefficiencies, particularly in high-throughput scenarios.

3D Reconstruction. cryo-EM has witnessed significant advancements in 3D reconstruction software, enabling near-atomic resolution of biological macromolecules. Among the most widely used tools are RELION, cryoSPARC, and EMAN2, each offering distinct advantages and limitations. RELION employs a Bayesian approach for iterative refinement, excelling in handling heterogeneous datasets and achieving high-resolution reconstructions¹². However, its computational intensity and steep learning curve often necessitate expert intervention. cryoSPARC, leveraging GPU acceleration and deep learning, significantly reduces processing time and is user-friendly¹³. EMAN2 provides a comprehensive suite for particle picking and 3D reconstruction but requires extensive manual tuning and lacks real-time feedback capabilities¹⁴. While these tools excel in specific aspects of 3D reconstruction, they remain confined to singular functionalities, lacking integration with upstream data acquisition or downstream atomic modeling. For instance, RELION and cryoSPARC require manual data transfer from microscope control software, introducing inefficiencies and potential errors. Moreover, none of these tools provide real-time quality assessment during data collection or seamless transition to atomic model building, resulting in fragmented workflows that hinder throughput and reproducibility.

The final computational stage in cryo-EM-based structural determination is atomic model building, where molecular coordinates are iteratively fitted into reconstructed density maps to elucidate atomic-level interactions.

While conventional model-building approaches employ physics-based energy minimization algorithms—incorporating electrostatic potentials, van der Waals interactions, and covalent bond geometry constraints—they remain heavily dependent on manual intervention for initial model placement, density interpretation, and stereochemical validation^15,16,17,18. This reliance on expert curation not only limits throughput but also introduces subjective biases, particularly in low-resolution regimes (< 4 Å) where density features are ambiguous.

To overcome these constraints, deep learning (DL)-based methodologies have emerged as transformative solutions, leveraging neural networks to automate feature extraction, model initialization, and refinement. These approaches demonstrate improved performance in handling noisy or incomplete density maps while significantly reducing manual intervention, as evidenced by recent benchmarks showing 40–60% reductions in user-dependent steps compared to conventional pipelines^19,20,21,22.

However, these methods invariably require exporting density maps from 3D reconstruction software and importing them into model-building systems. Prior to import, additional preprocessing steps—often involving third-party tools—may be necessary to optimize the density maps. This fragmented workflow necessitates repetitive data transfers and format conversions, creating inefficiencies. Moreover, if the initial model-building results are unsatisfactory, researchers must revert to the reconstruction stage for recalculation, followed by another round of model building and validation. Such iterative cycles significantly impede the throughput and operational efficiency of cryo-EM structural determination pipelines.

1.3 SMART: An Integrated Solution for End-to-End cryo-EM

Despite significant advancements in cryo-EM SPA, the lack of fully integrated computational solutions continues to impede its potential for high-throughput structural biology. Existing software packages, while excelling in automating isolated workflow components—such as data acquisition or 3D reconstruction—fail to bridge critical gaps between raw micrograph collection, density map refinement, and atomic model building. This fragmentation necessitates manual data transfer, format conversions, and redundant parameter tuning, resulting in inefficiencies that compound with dataset scale.

To address these limitations, we present Integrated cryo-EM workflow platform, an integrated computational platform unifying the entire cryo-EM SPA pipeline into a cohesive, AI-driven workflow. The integrated platform integrates three core modules: DataSmart (automated data acquisition), CryoSmart (hybrid neural network-enabled 3D reconstruction), and ModelSmart (deep learning-based atomic modeling).

The platform’s three computational modules are interconnected via a standardized data stream, enabling a streamlined data flow where output from one module can inform parameter adjustments in subsequent processing steps. This integration aims to reduce the number of manual steps and data transfers required in conventional cryo-EM workflows, potentially improving accessibility and efficiency for structural biology laboratories.

Protocol

The Expi293F cell line used for protein expression for human TRPML1 protein was purchased from Hunan Fenghui Biotechnology Co., Ltd (Catalog number: SC0236058). The identity of the cell line was properly authenticated via Short Tandem Repeat (STR) profiling (Supplementary File 1).

Ethical Statement
This study did not involve human participants, animals, or samples obtained directly from humans or animals. The human TRPML1 protein was produced by recombinant expression and purified in vitro. The Expi293F cell line utilized for recombinant protein expression is a commercially available, established immortalized cell line. Consequently, institutional ethical approval and written informed consent were not required.

NOTE: The Integrated cryo-EM workflow platform is structured into three core modules: an automated data acquisition module, a hybrid neural network-enabled 3D reconstruction module, and a deep learning-based atomic modeling module, with data serving as the communication interface between these software components, as illustrated in Figure 2. The Integrated cryo-EM workflow platform system adopts a Browser-Server (B/S) architecture, enabling users to perform all operations through a web interface, thereby ensuring usability and operational accessibility.

1. Automated Data Acquisition

NOTE: The Integrated cryo-EM workflow platform consists of three core modules: an automated data acquisition module, a hybrid neural network-enabled 3D reconstruction module, and a deep learning-based atomic modeling module. Data serves as the communication interface between these software components (Figure 2). The system uses a Browser-Server (B/S) architecture that enables all operations through a web interface.

1. Project Initialization
   1. Create a new project in an automated data acquisition module software platform.
   2. Select the option to inherit configuration settings from a recent project, if desired. Reuse the inherited parameters for electron microscope setup instead of creating a new configuration.
2. Beam Shift Inspection
   NOTE: Perform beam-shift adjustment within the transmission electron microscope (TEM) control software.
   1. Low-Magnification Check (Atlas and GridSquare)
      1. Open the Screening module. Acquire images at the Atlas and GridSquare magnifications.
      2. Verify that the images are properly illuminated and free of beam shift. Ensure that the image center is not positioned over the grid bars.
   2. Data Acquisition Magnification Check
      1. Acquire images at the data acquisition magnification (e.g., 60K or 100K). Evaluate beam alignment.
      2. Correct the beam shift before proceeding if image acquisition fails or significant beam shift is observed.
   3. Hole Eucentric Magnification Check
      1. Acquire images at the hole/eucentric magnification (typically 8K). Adjust the beam shift until the illumination spot is centered if the beam is not centered.
      2. Do not modify the aperture position. Verify that the aperture size matches the aperture used during data acquisition.
3. Z-Height Adjustment
   1. Auto-Eucentric Adjustment
      1. Navigate to the Screening page. Move the stage to a clean carbon film area at the hole/eucentric-height magnification. Open the Auto-Eucentric module on the AutoFunction page.
      2. Click Start. Monitor the log panel for feedback. Move to a different area and repeat the procedure if the operation fails.
   2. Auto-Focus Adjustment
      1. Open the AutoFocus module on the AutoFunction page. Click Start.
      2. Review the log messages. Adjust the parameters as indicated or repeat Section 1.3.1 if autofocus fails.
4. Magnification Alignment Check
   1. Alignment Inspection
      1. Locate a distinguishable feature at the Data magnification and acquire an image. Acquire an image at the Hole/EucentricHeight magnification.
      2. Compare the central regions of both images. Proceed to calibration if misalignment is observed between the Data and Hole magnifications.
      3. Inspect alignment between the Hole and GridSquare magnifications if the Data and Hole magnifications are aligned. Proceed without calibration if all magnification levels are aligned.
   2. Magnification Shift Calibration
      1. Navigate to the Calibration page. Open the Magnification Shift module. Begin with the Data magnification or the lowest misaligned magnification.
      2. Acquire a high-magnification image containing a distinct feature. Acquire a corresponding low-magnification image. Assess the displacement between the two images.
      3. Right-click the location corresponding to the feature center in the low-magnification image. Select Move Here. Reacquire the high-magnification image and verify alignment.
      4. Repeat the adjustment if necessary. Click Next to proceed to the next magnification pair after alignment is achieved. Repeat the calibration procedure for all remaining misaligned magnification pairs.
5. Preparation for Data Collection
   1. Perform atlas acquisition, square selection, square imaging, and hole selection
      1. Open the Atlas module. Leave the tile field blank to acquire all 81 tiles, or enter the desired number of tiles to acquire.
      2. Navigate to the Square Selection interface. Right-click to select suitable squares.
      3. Navigate to the Hole Selection page. Click Prepare All. Monitor the log window until image acquisition is complete.
      4. Select a captured square image from the image list. Use the Find Holes function to perform automatic hole selection when holes are clearly visible.
         CAUTION: Automatic hole detection may fail on grids with thick ice or surface contamination. Visually verify the detected holes before acquisition and use the semi-automated hole-selection method when automatic detection is unreliable.
      5. Use the Seed-Based Selection function when thick ice or contamination prevents reliable automatic selection (Figure 3).
         1. Select a central hole. Select two adjacent holes that form a right angle.
         2. Select two distant holes along the directions defined by the second and third holes.
         3. Click Submit. Enable Move All when processing another square. Drag one of the five seed holes to a clear hole. Click Submit.
   2. Template Definition
      1. Navigate to the Template Definition page. Click Acquire to obtain an image. Verify that the image contains a complete hole.
      2. Right-click the target area and select MoveStageTo if a complete hole is not visible. Reacquire the template image. Specify the imaging position and autofocus location by dragging the corresponding markers.
      3. Enter the number of frames per hole. Enter the desired defocus value (e.g., −1.5 µm). Verify all acquisition parameters (Figure 4).
6. Beam Shift Re-Check
   1. Repeat Section 1.2 after completing all preparation procedures.
7. Astigmatism and Coma Correction
   1. Astigmatism Correction
      1. Move the stage to a clean carbon film area. Adjust the height using AutoFocus to the desired defocus value (e.g., −1.5 µm). Open the Autostigmate module on the AutoFunction page.
      2. Click Start. Monitor the Ctffind output in the log panel. Review the astigmatism and resolution-fit values. Repeat the procedure in a different area if the astigmatism remains high.
   2. Coma Correction
      1. Open the Autocoma module if Autostigmate results remain suboptimal after multiple attempts. Click Start.
8. Automated Data Collection
   1. Navigate to the AutoCollect page. Click Start. Allow the system to perform AutoEmission and AutoZLP calibrations automatically.
   2. Acquire data using the parameters defined in the Screening, Data Acquisition, and Preparation settings. Monitor the automated collection process (Figure 5).
      PAUSE POINT: After data collection is complete, the acquired movie stacks can be stored and the protocol resumed at the image-processing stage at a later time without affecting downstream results.
9. Instant Remote Monitoring
   1. Monitor collection progress using the remote monitoring module. Review real-time motion-correction results generated using MotionCor2²³. Review real-time astigmatism and defocus measurements.
   2. Share the monitoring URL with authorized users. Access collection status and data-quality information from a computer or mobile device. Review an example of remote monitoring functionality in Figure 6.

2. Three-Dimensional Reconstruction

NOTE: Use 3D reconstruction module to perform three-dimensional reconstruction from previously acquired image datasets. The workflow includes project creation, micrograph import, contrast transfer function (CTF) estimation, particle picking, particle extraction, 2D classification, initial model generation, and 3D refinement (Figure 7).

1. Project Creation
   1. Navigate to the 3D reconstruction module homepage. Click Projects to open the project management page. Click NEW PROJECT.
   2. Enter the project name and specify the storage location. Click CREATE PROJECT. Click Open next to the newly created project. Click NEW Experiment to create a new experiment.
      NOTE: A project represents an overall data-processing session, whereas an experiment represents an individual processing workflow within a project. Multiple experiments can be created within a single project.
2. Import Files
   1. Click New Job. Select Import and click Create. Select the appropriate file type. Import the .mrc files generated during data acquisition. Enter the pixel size, acceleration voltage, and number of CPUs. Click Queue Job.
      NOTE: To terminate a running job, click Kill, then click Clear to return the job to the Building state. For completed or queued jobs, click Clear to reset the job status.
3. CTF Estimation
   1. Click New Job. Select CTF Estimation and click Create.
      NOTE: 3D reconstruction module supports multiple CTF-estimation algorithms. This protocol uses CTFFIND4.
   2. Click the Connect button for Exposures. Select the imported micrographs. Set the amplitude contrast to 0.1, minimum fitting resolution to 30 Å. maximum fitting resolution to 4 Å, minimum defocus search value to 5,000 Å, maximum defocus search value to 50,000 Å and defocus search step size to 500 Å.
   3. Click Queue Job.
4. Manual Particle Picking
   1. Click New Job. Select Particle Picking and then select Blob Picker. Click Create. Click the Connect button for Micrographs (Exposures).
   2. Select the imported micrographs. Enter the required parameters. Click Queue Job. Inspect 3–4 representative micrographs. Select particles using the left mouse button.
   3. Deselect particles using the right mouse button. Determine the particle diameter range from the selected particles. Click Done Picking! Extract Particles.
      NOTE: Measure particles from at least three micrographs.
5. Blob Picker
   1. Click New Job. Select Particle Picking and then select Blob Picker. Click Create. Connect the Micrographs (Exposures) input.
   2. Select the imported micrographs. Enter the required parameters. Click Queue Job.
6. Inspect Particle Picks
   1. Click New Job. Select Inspect Particle Picks. Click Create. Connect the Micrographs (Exposures) input. Connect the Particles input.
   2. Select the corresponding files. Enter the required parameters. Click Queue Job. Inspect 3–4 representative images after job completion.
   3. Verify correct particle selection. Click Done Picking! Output Locations.
7. Extract Particles from Micrographs
   1. Click New Job. Select Extract From Micrographs. Click Create. Connect the Micrographs (Exposures) input. Connect the Particles input. Select the corresponding files.
   2. Set the extraction box size to 2–3 times the particle diameter. Use an extraction box size of 400 pixels for TRPML1. Set the Fourier crop size to one-quarter to one-half of the extraction box size.
   3. Use a Fourier crop size of 100 pixels for TRPML1. Click Queue Job.
      NOTE: Reducing the Fourier crop size decreases GPU computational demand.
8. Two-Dimensional Classification
   1. Click New Job. Select 2D Classification. Click Create. Connect the Particles input.
   2. Import the particle stack. Enter the required parameters. Click Queue Job.
      NOTE: Select 50–200 classes. Use fewer classes for smaller datasets and more classes for larger datasets.
9. Select Two-Dimensional Classes
   1. Click New Job. Select Select 2D Classes. Click Create. Connect the Particles input. Connect the Templates input.
   2. Import the required files. Click Queue Job. Wait until the job enters the Waiting state. Click Waiting.
   3. Select clear and well-defined 2D classes. Click Done.
10. Initial Model Generation
    1. Click New Job. Select Abinit Reconstruction. Connect the Particles input. Import the selected particles.
    2. Enter the number of initial classes. Enter the symmetry parameter. Click Queue Job.
11. Heterogeneous Refinement
    NOTE: Use heterogeneous refinement for heterogeneous samples and homogeneous refinement for homogeneous samples.
    1. Click New Job. Create a Heterogeneous Refinement job. Connect the Particles input. Connect the Volume input.
    2. Select the corresponding files. Enter the refinement parameters. Specify the appropriate symmetry (e.g., C1). Click Queue Job.
12. Homogeneous Refinement
    1. Click New Job. Create a Homogeneous Refinement job. Connect the Particles input. Connect the Volume input.
    2. Optionally connect the Mask input. Import the particle stack and the best-performing volume class from the previous refinement step. Enter the required parameters.
    3. Specify the appropriate symmetry (e.g., O or C1). Click Queue Job. Click Output after job completion. Inspect the reconstructed density map in the browser interface (Figure 8).
13. Map Enhancement
    NOTE: Map Enhancer uses DeepEMhancer²⁴. Use half-maps whenever possible. Post-translational modifications and ligand molecules were not included in the DeepEMhancer training set and may not be enhanced accurately. Map Enhancer performs local sharpening, automated masking, and denoising.
    1. Click New Job. Create a Map Enhancer job. Import the Volume generated during Homogeneous Refinement.
    2. Import the corresponding Mask. Import map files through the Map Data Path option if required. Execute the job.

3. A Deep Learning-Based Atomic Modeling

NOTE: A deep learning-based atomic modeling module is based on the previously developed SMARTFold framework²⁵ and supports atomic model building for proteins and nucleic acids (RNA/DNA) from cryo-EM density maps.

1. Launch a deep learning-based atomic modeling module
   1. Log in to a deep learning-based atomic modeling module system. Open the task submission interface. Upload the cryo-EM density map in the left panel.
   2. Enter the required input parameters. Submit the modeling task.
2. Upload the Cryo-EM Density Map
   1. Upload the density map in .mrc, .ccp4, or .map format. Verify that the file size is less than 100 MB.
      NOTE: If the density map exceeds 100 MB, reduce the file size before uploading.
      1. Crop the density map to the region of interest using UCSF ChimeraX²⁶ with the command: volume zone. Resample the map to a larger pixel size.
      2. Apply a soft mask to remove solvent regions.
         NOTE: Inspect the density map before uploading to ensure optimal model performance and reduce the risk of processing errors.
3. Inspect and Prepare the Density Map
   1. Perform Density Map Cropping. For this Crop the density map to remove unnecessary regions. Verify that the cropped map contains only the region required for model building.
      NOTE: Cropping reduces file size and helps prevent memory overflow during model generation.
   2. Then check for Mirror Inversion. Inspect the density map for possible mirror inversion. Examine α-helical regions within the map. Verify that α-helices exhibit a right-handed configuration and spiral upward in a counterclockwise direction. Flip the density map if a mirrored structure is detected.
4. Upload the Sequence File
   1. Upload the corresponding sequence file. Alternatively, enter the sequence manually. Provide all protein or nucleic acid sequences included in the reconstruction.
   2. Specify heteromeric or homomeric sequence information when applicable. Enter the Protein Name, if desired (Optional Step).
5. Specify the Map Resolution. Enter the reported resolution of the cryo-EM density map. Select the Computing Partition. Select the GPU computing node to be used for model generation.
6. To specify CPU Resources, enter the number of CPUs to be allocated. Do not exceed eight CPUs. Review all uploaded files and parameters. Click SUBMIT to initiate model generation.
7. Inspect the Predicted Atomic Model
   1. Open the interactive visualization window after job completion. Rotate the model to inspect structural features. Zoom in and out to examine local and global structural details.
   2. Inspect the model from multiple orientations. Evaluate model quality, spatial conformation, and atomic-level features. Use the generated model for downstream analysis and validation.
   3. Refer to Figure 9 for an example of the visualization interface.

Results

TRPML1, also known as mucolipin-1 (MCOLN1), is a lysosomal cation channel essential for maintaining ionic homeostasis, regulating vesicular trafficking, and supporting autophagy in mammalian cells^27,28. Dysfunction of TRPML1 due to genetic mutations has been directly linked to mucolipidosis type IV, a severe neurodegenerative lysosomal storage disorder characterized by abnormal neurodevelopment, retinal degeneration, and iron-deficiency anemia^29...

Discussion

The structural determination of biological macromolecules via cryo-electron microscopy (cryo-EM) inherently involves a protracted, multi-stage workflow requiring tight integration of computational tools and reliance on expert intervention. The integrated platform addresses this challenge by unifying data acquisition, 3D reconstruction, and atomic modeling into a single, vertically integrated framework—a step toward more streamlined cryo-EM workflows. By embedding AI-driven algorithms across the pipeline (e.g. geome...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
CryoSmart	Shuimu BioSciences Ltd.	V 1.0	3D reconstruction module of the SMART platform; hybrid neural network-enabled
Cryo-electron microscope	Shuimu BioSciences Ltd.	SHUIMU TOTEM 300S	Used for data acquisition; equipped with energy filter
DataSmart	Shuimu BioSciences Ltd.	V 1.0	Automated data acquisition module of the SMART platform
DeepEMhancer	Open source	V0.17	Deep learning-based cryo-EM volume post-processing; integrated within CryoSmart Map Enhancer
Direct electron detector	Gatan	GATAN K3 (6k x 4k)	Camera used for movie recording
Grid	Quantifoil	Au R1.2,1.3	Cryo-EM specimen grid
ModelSmart	Shuimu BioSciences Ltd.	V 1.0	Deep learning-based atomic modeling module of the SMART platform; extends the SMARTFold framework
SMART Platform	Shuimu BioSciences Ltd.	V 1.0	Integrated cryo-EM platform; Browser-Server (B/S) architecture; includes DataSmart, CryoSmart, and ModelSmart modules
TEM control software	Shuimu BioSciences Ltd.	V1.0	Electron microscope control software for beam shift adjustment
Topaz	Open source	V 2.5	Deep learning-based particle picking tool; integrated as plug-in within CryoSmart
TRPML1 protein sample	Shuimu Biosciences Ltd.	SMPP-0053	Human TRPML1 (mucolipin-1); prepared in-house by Protein and Crystallography Department
Vitrification device	Thermo Fisher Scientific	Vitrobot Mark IV	For cryo-EM sample preparation
UCSF Chimera X	UCSF, USA	https://www.cgl.ucsf.edu/chimerax/	General purpose software for display, analysis and more
SMART (Shuimu Automated Reconstruction Technology)	Shuimu BioSciences Ltd.	N/A	Integrated cryo-EM workflow platform
SMARTFold	Shuimu BioSciences Ltd.	N/A	Deep-learning framework underlying ModelSmart
Leginon	National Resource for Automated Molecular Microscopy (NRAMM)	RRID:SCR_016731	Automated cryo-EM data acquisition software
SerialEM	University of Colorado	RRID:SCR_017293	Automated TEM/cryo-EM acquisition software
EPU	Thermo Fisher Scientific	N/A	Automated cryo-EM data collection software
EPU Multigrid	Thermo Fisher Scientific	N/A	Multi-grid automated cryo-EM acquisition
Gatan Latitude	Gatan Inc.	N/A	Electron microscopy acquisition software
JADAS	JEOL Ltd.	N/A	Automated cryo-EM acquisition system
Auto-EMation	Tsinghua University	N/A	Automated cryo-EM acquisition software
RELION	MRC Laboratory of Molecular Biology	RRID:SCR_016274	Bayesian cryo-EM reconstruction software
cryoSPARC	Structura Biotechnology Inc.	RRID:SCR_016501	Cryo-EM image processing and reconstruction
EMAN2	Baylor College of Medicine	RRID:SCR_017270	Cryo-EM image processing suite
CTFFIND4	MRC Laboratory of Molecular Biology	RRID:SCR_016732	CTF estimation software
Gctf	MRC Laboratory of Molecular Biology	RRID:SCR_016500	GPU-accelerated CTF estimation
MotionCor2	Howard Hughes Medical Institute / UCSF	RRID:SCR_016499	Motion correction software
DeepEMhancer	Spanish National Center for Biotechnology (CNB-CSIC)	RRID:SCR_021214	Deep-learning cryo-EM map enhancement
Topaz	University of California Berkeley	RRID:SCR_018037	Deep-learning particle picking
UCSF ChimeraX	University of California San Francisco	RRID:SCR_015872	Molecular visualization software
CTFFIND4	Grigorieff Lab	RRID:SCR_016732	CTF fitting used in CryoSmart workflow
EMDB-8840	Electron Microscopy Data Bank	Accession: EMD-8840	Human TRPML1 reference map
EMPIAR-10204	Electron Microscopy Public Image Archive	Accession: EMPIAR-10204	β-galactosidase benchmark dataset
Transmission Electron Microscope (TEM)	Not specified	N/A	Used for cryo-EM data acquisition
FEI Tecnai Electron Microscope	FEI (now Thermo Fisher Scientific)	Not specified	Mentioned in cited workflow
Energy Filter	Not specified	N/A	Used during automated collection
GPU Computing Node	Not specified	N/A	Used for ModelSmart calculations