AutoLEI: An XDS-based pipeline with graphical user interface for automated real-time and offline batch 3D ED/microED data processing

Three-dimensional electron diffraction (3D ED), also known as microcrystal electron diffraction (microED), is an emerging method for determining structures from submicron-sized crystals. With the development of rapid and convenient data collection protocols, acquiring dozens of datasets in a single 3D ED/microED session has become routine. A fast and automated workflow for processing, scaling and merging a large number of 3D ED/microED datasets can significantly accelerate the structure determination process. Herein, we present an XDS-based pipeline with a graphical user interface for automated real-time and offline batch 3D ED/microED data processing (AutoLEI). We demonstrate the functionality and applications of the pipeline through four examples, using both offline and real-time data processing capabilities. The samples include small organic molecules, metal–organic frameworks (MOFs) and proteins, showcasing the versatility and efficiency of AutoLEI in various applications.

1. Introduction

Three-dimensional electron diffraction (3D ED), also known as microcrystal electron diffraction (microED), has been developed into a robust method for crystal structure determination over the last few decades (Kolb et al., 2007; Zhang et al., 2010; Shi et al., 2013; Gemmi et al., 2019). The method enables rapid and accurate structure determination using commercially available transmission electron microscopes (TEMs). A wide range of structures, including zeolites (Guo et al., 2015; Cho et al., 2023; Lu et al., 2024), metal–organic frameworks (MOFs) (Huang, Willhammar & Zou, 2021; Huang, Grape et al., 2021), pharmaceuticals (van Genderen et al., 2016; Lightowler et al., 2022) and macromolecules (Shi et al., 2013; Xu et al., 2019; Martynowycz et al., 2022), have been successfully solved with 3D ED/microED.

Benefiting from the strong interaction between electrons and matter, 3D ED/microED data can be collected from crystals that are too small for single-crystal X-ray diffraction (Tremlett et al., 2025). An ideal TEM grid typically contains hundreds of diffracting crystals or crystal fragments. With the latest 3D ED/microED protocol, it is possible to determine structures of high-symmetry stable crystals through a single dataset within a few minutes (Cichocka et al., 2018; Plana-Ruiz et al., 2020; Ito et al., 2021; Unge et al., 2024). However, owing to the inherent limitations of the goniometer, data merging for crystals of low symmetry is usually required to achieve high data completeness. For highly beam sensitive material, a small-wedge data collection and merging strategy can also be employed to increase the signal-to-noise ratio and data completeness (Clabbers et al., 2021). Furthermore, selectively merging 3D ED/microED datasets can improve data quality by increasing redundancy (Xu et al., 2018; Samperisi et al., 2021; Bengtsson et al., 2022). More recently, high-throughput data collection has been achieved using software such as Instamatic (Cichocka et al., 2018), Automated EM Data Acquisition with SerialEM (de la Cruz et al., 2019), Leginon (Cheng et al., 2021), EPUD (Thermo Fisher Scientific, TFS) and Latitude D (GATAN) or a dedicated electron diffractometer, such as Synergy-ED (Rigaku) and ELDICO ED-1 (ELDICO Scientific). It is feasible to perform qualitative phase analysis by collecting and processing well populated datasets (Luo et al., 2023; Unge et al., 2024; Lightowler et al., 2024). In all the above cases, an efficient, automated and even real-time data processing solution is highly desirable to enhance the efficiency of structure determination and phase analysis using 3D ED/microED.

A large selection of well established software, such as X-ray detector software (XDS) (Kabsch, 2010), PETS2 (Palatinus et al., 2019), CrysAlisPro-ED (Truong et al., 2023), xia2.multiplex (Gildea et al., 2022) and DIALS (Winter et al., 2018, 2022), is available for processing 3D ED/microED data. Data processing can be performed via either command-line instructions or graphical user interfaces (GUIs) (Powell et al., 2021; Brehm et al., 2023), offering flexibility to users. However, format conversion, sorting, processing, analyzing and merging a large number of 3D ED/microED datasets can be time consuming and laborious. Automated batch processing and real-time processing capabilities are still underdeveloped for 3D ED/microED. In our previous work, Edtools (Wang et al., 2019) was developed to process batch datasets in offline mode, which only works in a command line. A processing pipeline Scipion-ED (Bengtsson et al., 2022) embedded in Scipion was also developed. However, the complex operation and specialized platform requirements make these programs difficult for beginners to use.

In this work, we utilized XDS as the data processing engine and developed a pipeline with a graphical user interface for automated real-time and offline batch 3D ED/microED data processing (AutoLEI). Based on the fast data processing performance of XDS, AutoLEI enables both real-time and offline batch 3D ED/microED data processing. This Python-based pipeline is straightforward to install and provides users with the options and parameters that are essential for working with electron diffraction data. Meanwhile, data processing statistics and quality indicators are also optimized for 3D ED/microED. Here, we demonstrate how to use AutoLEI to process data collected from different types of crystals, including an organic molecule, a metal–organic framework and two proteins, on different microscopes and detectors. An example of real-time data processing using AutoLEI is also presented. Furthermore, a detailed tutorial with training datasets is provided.

2. AutoLEI implementation

2.1. AutoLEI installation and supported data formats

AutoLEI is an open-access, Python-based 3D ED/microED data processing pipeline, available through PyPI, GitLab and Zenodo. The software operates on Linux systems or within the Windows Subsystem for Linux (WSL) environment with Python version 3.8 or greater, and requires XDS to be pre-installed and configured globally. AutoLEI also provides a one-click installation command that automatically installs all required dependencies on GitLab. Detailed installation instructions are available on corresponding GitLab wiki (see Data availability). In version 1.0.0, the native image format is SMV for XDS data processing. AutoLEI can convert MRC images collected by EPUD (TFS), and TIFF images collected on CheeTah (Amsterdam Scientific Instruments, ASI), OneView (GATAN) and Timepix hybrid pixel detectors (ASI). Other formats need to be converted in advance to SMV format using other software, for example REDp (Wan et al., 2013) and RosettaSciIO (Prestat et al., 2025). Real-time data processing is designed for raw MRC image data collected using EPUD (TFS) as well as data collected using Instamatic (Cichocka et al., 2018). Future updates are expected for improved compatibility and increased support, while users can also add new functionality through the programming interface.

2.2. AutoLEI data processing workflow

The primary aim of AutoLEI is to process and analyze a batch of 3D ED/microED datasets using XDS offline or in real time. AutoLEI builds upon the experience from our earlier software, Edtools (Wang et al., 2019). While Edtools relies on command-line input and requires an existing XDS.INP file generated by Instamatic (Cichocka et al., 2018), AutoLEI introduces a graphical user interface (GUI) and supports image data collected using various acquisition software. Moreover, AutoLEI's workflow has been specifically optimized for rotational electron diffraction data, allowing the processing to be significantly accelerated and enabling efficient real-time analysis. Fig. 1 shows the workflow of AutoLEI. Data processing goes through the steps of XDSRunner, CellCorr, XDSRefine, MergeData and Cluster&Output. These steps are implemented as tabs in the AutoLEI interface, as shown in Figures S1–S8 in the supporting information. By following this workflow, final HKL files are obtained for downstream structural determination. Videos 1–3 in the supporting information showcase the offline data processing of three examples: tyrosine, MOF SU-100 and the protein H. sapiens MutT homolog 1 (MTH1) while Video 4 shows the real-time data processing of triclinic lysozyme.

Figure 1 An overview of the AutoLEI workflow.

2.2.1. Input

Input allows users to set up the work directory, instrument parameters and measurement settings (Figure S1). The instrument parameters can be loaded from a predefined instrument model or existing input files. Alternatively, they can be given manually when datasets collected from different instruments and detectors are used. Input creates the Input_parameters.txt file (see an example in Section S3.2 of the supporting information), which will be used for generating input files XDS.INP for both real-time and offline 3D ED/microED data processing.

2.2.2. XDSRunner

XDSRunner generates the corresponding XDS.INP files for each dataset (Figure S2). It then launches XDS to process these datasets. No prior knowledge of the sample is required. AutoLEI processes all datasets sequentially under the work directory and parses statistics from XDS outputs. After data processing, AutoLEI will generate a summary table named xdsrunner.xlsx. The table contains key parameters and quality indicators such as space group (SG), unit-cell parameters (Unit cell), cell volume (Vol.), indexing rate (Index%), I/σ(I)^asymptotic (ISa), redundancy-independent merging R factor (R_meas), correlation between reflection intensities of random half datasets (CC_1/2), completeness and estimated resolution (Reso.) of each dataset for inspection (Diederichs, 2010; Karplus & Diederichs, 2012). SG, unit cell and ISa are read from XDS outputs directly while others are calculated by AutoLEI based on XDS HKL files. The high-resolution cut-off of each dataset is estimated by checking CC_1/2, R_int and the signal-to-noise ratio of each resolution shell.

XDSRunner also contains supporting functions as described below:

Format converter: Converts raw image files to SMV format and writes the metadata into SMV files.

Find beam center: Finds the beam center automatically with or without a beam stop.

Estimate symmetry & cell-cluster: Clusters data based on unit-cell parameters and estimates the Laue group. The Laue group is estimated based on R_meas and CC_1/2 using four indicators. The unit-cell clustering is performed based on a distance matrix. These ensure that data are processed in the correct Laue group. Details of these functions can be found in Section S1.1 of the supporting information.

2.2.3. CellCorr

CellCorr allows users to run XDS with given unit-cell parameters and space group (Figure S3). It allows re-indexing of the original HKL file using the space group and unit-cell parameters suggested by either Estimate symmetry & cell-cluster in XDSRunner or determined by other software packages such as REDp (Wan et al., 2013) or PETS2 (Palatinus et al., 2019). This ensures that the datasets will be processed, scaled and merged in the correct Laue group and unit cell. CellCorr outputs a summary table named xdsrunner2.xlsx containing the same data quality indicators as xdsrunner.xlsx.

2.2.4. XDSRefine

XDSRefine provides the opportunity to update XDS keywords in the input files for single, selected or all datasets if users are not satisfied with the outputs from CellCorr (Figure S4). The step is useful for problematic data that cannot be processed using default settings. In AutoLEI version 1.0.0, XDSRefine also provides automatic refinement of the rotation axis, and excludes problematic frames defined by outliers; see Section S1.2 for more details. The reconstructed reciprocal lattice of each dataset can also be visually inspected. XDSRefine updates xdsrunner2.xlsx and the table will then be used for downstream data scaling and merging.

2.2.5. MergeData

MergeData analyzes statistics in xdsrunner2.xlsx and makes initial selections of datasets for merging (Figure S5). Users can then inspect key indicators listed in the table and apply a filter for scaling/merging. The default value will be provided as a reference. The information about the filtered data is written to the file xdspicker.xlsx. MergeData will then initiate XSCALE and output scaled data as an HKL file.

2.2.6. Cluster&Output

Cluster&Output performs cluster analysis based on pairwise correlation coefficients of reflection intensities in the file XSCALE.Lp generated from the last step (Figure S6). A dendrogram is produced for visual inspection for further clustering (Wang et al., 2019). Cluster&Output will then perform scaling/merging of each cluster, outputting corresponding HKL files and data processing statistics (Figure S7). Web-page-based reports will also be generated for inspection of the data reduction.

2.3. Real-time processing

Real-time 3D ED/microED data processing is a fully automated pipeline in AutoLEI (Figure S8). With default or customized settings, it provides crucial feedback to users for data quality analysis, evaluates data collection parameters and generates result HKL files in real time. The setup of the real-time processing function is found on the RealTime tab of AutoLEI, as shown in Fig. 2. Real-time processing reads Input_parameters.txt from Input and continuously monitors the input folder into which the data are being saved. Once a single dataset collection is completed, real-time processing will try to process the data in either screening mode or merging mode:

Figure 2 RealTime microED work page. (a) Screening mode. (b) Merging mode. `Iters' in (b) denotes the number of merging iterations.

(a) Screening mode: When the unit-cell parameters and space group are not provided, the data are processed without performing data merging.

(b) Merging mode: When the unit-cell parameters and space group are provided, the data will be processed in the specified cell and space group. If the data quality satisfies the filtering condition, it will be assigned as a valid dataset and merged with other valid datasets.

In the RealTime tab, estimated resolution, CC_1/2 and ISa for individual data sets or merged data are plotted and updated live, as shown in Figs. 2(a) and 2(b), respectively. Real-time processing currently supports the data collected using EPUD (TFS) and Instamatic (Cichocka et al., 2018).

3. Examples

To demonstrate the functions and applications of AutoLEI, we chose four different samples (Fig. 3): an organic compound, tyrosine (P2₁2₁2₁, a = 6.92 Å, b = 21.15 Å, c = 5.84 Å) (Boggs & Donohue, 1971), a bis­muth metal–organic framework, SU-100 (I2/a, a = 17.85 Å, b = 9.61 Å, c = 21.05 Å, β = 96.77°) (Grape et al., 2020), a protein, MutT homolog 1 (MTH1) (P2₁2₁2₁, a = 59.5 Å, b = 67.1 Å, c = 80.1 Å) (Svensson et al., 2011), and triclinic lysozyme (P1, a = 27.07 Å, b = 31.25 Å, c = 33.76 Å, α = 88.0°, β = 108.0°, γ = 112.1°) (Wang et al., 2007). 3D ED/microED data were collected on a variety of microscopes and detectors using different software and processed using AutoLEI, as summarized in Table 1. Among them, tyrosine (Video 1), SU-100 (Video 2) and MTH1 (Video 3) use the offline data processing protocol, while lysozyme (Video 4) showcases the online data processing capability. The sample preparation details are given in Section S2.

Table 1
Summary of data used for demonstrating AutoLEI processing

JEOL: Japan Electron Optics Laboratory Co., Ltd. ASI: Amsterdam Scientific Instruments. TFS: Thermo Fisher Scientific.

Sample	Microscope	Detector	Data collection software	Total datasets	Data collection time	Data processing time†
Tyrosine (P212121)	JEM-2100 LaB6 (JEOL)	Timepix hybrid pixel detector (ASI)	Instamatic	12	∼1 h 20 min	∼5 min
SU-100 (I2/a)	JEM-2100 LaB6 (JEOL)	Timepix hybrid pixel detector (ASI)	Instamatic	16	∼2 h 15 min	∼7 min
MTH1 (P212121)	Krios G3i (TFS)	Ceta-D (TFS)	EPUD (TFS)	38	∼2 h 14 min	∼12 min
Lysozyme (P1)	Krios G2 (TFS)	Ceta-D (TFS)	EPUD (TFS)	72	∼6 h	real time (< 1 min)

†Timing is from the start of data processing to obtaining usable HKL files (Videos 1–3).

Figure 3 Structures used for the demonstration of AutoLEI. (a) L-tyrosine obtained using a selected high-quality dataset. The electrostatic potential map (in blue, contoured at 1 RMSD) and difference map [positive in green and negative in red, contoured at 3 RMSD drawn in COOT (Emsley et al., 2010)] are superimposed. (b) The framework of the bismuth-containing MOF SU-100 viewed along the b axis (BiO7 polyhedra shown in purple). (c) Apo-MTH1. (d) Lysozyme model near a nitrate anion with no-filled 2Fo − Fc electrostatic potential map (blue, contoured at 1.5 RMSD) superimposed.

3.1. Rapid data quality assessment and selection of a high-quality dataset for the structure determination of tyrosine

Because of the rapid 3D ED/microED data collection, many datasets can often be collected from different crystals. It is therefore important to make a rapid data quality assessment, in particular to identify datasets with the highest and lowest quality. Here we use tyrosine crystals as an example to demonstrate the capability of AutoLEI.

Tyrosine is a naturally occurring amino acid and an essential protein building block. Owing to its high orthorhombic symmetry and relatively high stability under the electron beam, it is possible to determine the structure of tyrosine crystals using a single 3D ED/microED dataset collected over a large tilt range. A total of 12 tyrosine datasets were collected by Instamatic on a JEOL JEM2100 microscope at 200 kV using a Timepix detector. The crystal tracking function was used to ensure that most of the data were collected over a large tilt range (∼130°) (Cichocka et al., 2018).

In the XDSRunner tab, XDS first ran without a specified space group and automatically determined the Laue group and Bravais lattice. A default resolution range from 20 to 0.8 Å was set for the initial data processing. Unit-cell parameters, space group, ISa, CC_1/2, completeness and estimated resolution were extracted from XDS outputs, as shown in Fig. 4(a). Most of the datasets exhibit the estimated resolution of 0.8 Å, which equals the pre-defined high-resolution limit. This suggests that the resolution range should be extended. Notably, XDS assigned the space group P222 (No. 16) to 10 out of the 12 datasets while giving wrong space groups for two datasets (No. 7 and 12). By applying the built-in Estimate symmetry & cell clustering function – optimized for electron diffraction data – it was confirmed that these two datasets share the same Laue group and have similar unit-cell parameters to the other datasets. This allows new XDS runs to be performed in the CellCorr tab with the specified space group (P222) and unit cell (a = 6.000 Å, b = 7.027 Å, c = 21.718 Å, α = β = γ = 90°) [Fig. 4(a)]. Next, we applied XDSrefine to refine the rotation axis, divergence and mosaicity, remove scale outliers, and change the resolution range to 30–0.5 Å, as indicated in Fig. 4(b). After the refinement, statistic values such as CC_1/2 and estimated resolution are improved for most of the datasets. For example, the resolution of experiment 5 is improved from 1.0 to 0.74 Å, where the CC_1/2 increases from 91.78% to 96.11%.

Figure 4 Batch data processing of 12 L-tyrosine datasets using AutoLEI. (a) Result table after running XDSRunner. The suggested unit-cell parameters with specific Laue group and lattice type setting were extracted from the function Estimate Symmetry & Cell Cluster. (b) Result table after running XDSRefine. View Reciprocal Space was designed to inspect the reciprocal space of a single dataset.

The choice of optimal datasets can vary depending on which key indicators are prioritized. From our experience, we recommend datasets with ISa > 5, CC_1/2 > 95% and completeness > 80% as being promising for single-data-set-based structure determination. Here we selected experiment 10 because it has the highest completeness (87.61%) and also a high ISa (7.51), CC_1/2 (98.18%) and resolution (0.68 Å). AutoLEI also enables the systematic absences in reciprocal space to be visually inspected. By rapidly checking the reciprocal lattices of all datasets using AutoLEI, the presence of systematic absences could be identified on h00, 00l in experiment 10 and 0k0 in experiment 12 (Figure S9). The space group was deduced as P2₁2₁2₁. The time used for data processing and analysis was approximately 5 minutes in total. The integrated data were then used for structure solution and kinematic refinement. The refinement converged to a final R₁ of 11.90% for 1253 reflections with I > 2σ(I) and 162 parameters (Table S1). All hydrogen atoms, including the proton near the N atom, could be found as shown in Fig. 3(a). The structure could also be solved and refined using other datasets (Table S1) except for two datasets (experiment 3 and experiment 7) that had too low completeness (43.19% and 5.64%, respectively) [Fig. 4(a)] due to beam blockage by the grid.

In conclusion, AutoLEI can directly generate data information tables so that users can make their own assessment of the data quality. This can help users to identify the best datasets for structure determination or to decide whether data merging is needed in cases where the completeness is low. It can also help to eliminate low-quality datasets.

3.2. Interactive clustering and merging of multiple datasets from a monoclinic bis­muth metal–organic framework, SU-100

For crystals with low symmetry, merging datasets collected from crystals from different orientations is often needed to achieve high completeness. In addition, data merging can minimize the effects of dynamical scattering and improve the data quality (Xu et al., 2018). Here we chose a monoclinic bis­muth metal–organic framework (I2/a, a = 17.85 Å, b = 9.61 Å, c = 21.05 Å, β = 96.77°) (Grape et al., 2020). In this study, a total of 16 datasets were collected using Instamatic, and all data were processable after XDSRefine, as shown in Fig. 5(a). The data resolution ranges from 0.54 (experiment 8) to 0.78 Å (experiment 11), and the completeness ranges from 20.82% (experiment 7) to 82.61% (experiment 2). Fourteen out of the 16 datasets have an ISa higher than 5 and were then merged during the XDSRefine step after specifying the space group and unit-cell parameters [Fig. 5(b)]. We further conducted cluster analysis based on the pairwise correlation coefficients of reflection intensities (Wang et al., 2019). As shown in Fig. 5(c), a threshold of 0.5 was selected that gives a significant distance between the four clusters. The final merged HKL file from 11 datasets was obtained in 7 minutes. The resolution cut-off of the merged data estimated by AutoLEI was 0.62 Å. All non-H atom positions could be directly located and refined anisotropically using the merged data. The refinement converged to an R₁ of 19.46% for 4231 unique reflections with I > 2σ(I) and 200 parameters, as shown in Table S2.

Figure 5 Data processing, analysis and merging of datasets of SU-100 by AutoLEI. (a) XDSRefine work page showing the result table of 16 datasets. (b) MergeData work page. Fourteen out of 16 datasets with ISa > 5 were used for initial merging after several clicks. (c) Cluster&Output work page. Clusters were created based on pairwise reflection intensity correlations and a final HKL file for structure determination was prepared that contained merged data from the largest cluster of 11 datasets (Wang et al., 2019).

It is important to mention that data merging can significantly improve the data quality (Xu et al., 2018; Samperisi et al., 2021). We compared the structure refinement using the merged dataset and a single dataset from experiment 2, which has the highest completeness of all single datasets, as shown in Figure S10. While the structure could be refined anisotropically using the merged data, anisotropic refinement using a single dataset of experiment 2 gave non-positive-definite atomic displacement parameters for 19 atoms.

3.3. Processing small-wedge 3D ED/microED data from the protein MutT homolog 1 (MTH1)

For structure determination of very beam sensitive crystals, such as protein crystals (Shi et al., 2013), a small-wedge data collection strategy (Clabbers et al., 2021) can be used, where a large number of 3D ED/microED datasets are collected from different crystals, each with a small tilt range (10–20°) and a low completeness. AutoLEI is ideal for conducting repetitive data processing, data scaling and merging of such batch data.

MTH1 plays an important role in sanitizing oxidized dNTPs from the free nucleotide pool, preventing their incorporation into DNA, which reduces genotoxicity (Yoshimura et al., 2003; Gad et al., 2014; Jemth et al., 2018). Here, we used apo-MTH1 as an example to demonstrate how small-wedge 3D ED/microED data can be effortlessly processed and merged with AutoLEI. A total of 38 datasets were collected using EPUD (TFS) and the tilt range was limited to 15 or 20° in order to collect data with a high signal-to-noise ratio before the crystals were damaged by the electron beam. Only 28 out of 38 datasets were processable after XDSRefine. Fig. 6(a) shows the dendrogram of the clustering analysis after the MergeData step in AutoLEI. A stepwise clustering analysis using different distance thresholds can be performed in AutoLEI to find the optimal one. As shown in Fig. 6(a), five thresholds were tested, resulting in five groups containing 27, 26, 23, 18 and 13 datasets, respectively. Each group is called disX-clsY, where X is the threshold and Y is the main cluster under the threshold. The first four groups showed comparable completeness (74.93–75.66%), while the last group had a slightly lower completeness (70.23%). We chose dis4-cls1 for data merging because it has the highest CC_1/2 and lowest R_meas among the first four groups. A web-page-based report was then generated and used for data reduction of dis4-cls1. Fig. 6(b) indicates the relationships between CC_1/2, R_meas and completeness against resolution drawn from the report of dis4-cls1. The estimated resolution was 2.76 Å and it took less than 12 minutes to obtain the final HKL file using AutoLEI. The structure of MTH1 could be solved by Phaser (McCoy et al., 2007) and refined using Phenix.refine (Afonine et al., 2012; Liebschner et al., 2019). The final R_work and R_free converged to 21.07% and 26.82%, respectively, as shown in Table S3.

Figure 6 Data processing, analysis and merging of 38 MTH1 datasets by AutoLEI. (a) Clustering analysis after data merging in the AutoLEI interface. A total of 5 cluster selections were made based on different distances. A summary of the key statistics of these 5 clusters was tabulated in the command window. (b) Web-page-based report plots of Cluster_1 based on threshold 4 (18 datasets in total). Plots were captured from the report. From top to bottom: CC1/2 versus resolution, Rmeas and Rint versus resolution, and completeness versus resolution.

3.4. Real-time batch data processing for triclinic lysozyme

Real-time data processing for single-crystal X-ray diffraction is implemented at most synchrotron X-ray facilities. This capability is also desirable for 3D ED/microED, as it provides live processing statistics on the current batch data collection. This allows users to inspect the data quality in order to decide how to optimize the data collection strategy and to stop data collection when enough data are collected. By combining AutoLEI with automated data collection protocols, rapid structure determination and phase analysis become possible. Here, we used triclinic lysozyme crystals to demonstrate the real-time 3D ED/microED data processing capability of AutoLEI.

Triclinic lysozyme crystals diffract to a very high resolution (Wang et al., 2007). However, triclinic lysozyme is challenging for 3D ED/microED because of its low symmetry and preferred crystal orientation. Here, a small-wedge data collection strategy was employed to increase the data resolution. The tilt range of individual crystals was restricted to 15° to keep beam damage minimal. To achieve real-time data merging (merge mode), the unit-cell parameters and space group need to be given as prior knowledge. These two input parameters can easily be obtained by first collecting a high-tilt 3D ED/microED dataset. Data with both CC_1/2 higher than 90% and ISa higher than 5 were considered to be good and subsequently used for real-time merging iterations. These parameters could also be customized by editing the strategy file settings. Furthermore, the resolution of each good dataset as well as the statistics of merged data completeness and CC_1/2 were updated live. Users can then use this information to check whether the current data collection strategy or data filtering strategy should be modified.

A total of 71 datasets were collected on lysozyme crystals in 6 h. The completeness of the data reached 95.8%, while the CC_1/2 was 98.68% at the resolution cut-off of 1.3 Å, as shown in Fig. 7(a). Owing to preferred orientation, the completeness only reached 50% and 78% when merging the first 10 and 21 datasets, respectively. In order to reach a higher completeness, we had to change the data collection strategy and look for crystals with different morphologies on the TEM grids. The data collection was terminated after the merged data reached a completeness higher than 95% at the pre-set resolution cut-off of 1.3 Å and a CC_1/2 above 98%. The web-page-based report was then generated to estimate the real high-resolution cut-off, which was set at 1.1 Å at the CC_1/2 of 30%, as shown in Figs. 7(b) and 7(c). The final HKL file merged from 56 datasets was used immediately for structure solution using Phaser (McCoy et al., 2007) and refinement using Phenix (Afonine et al., 2012; Liebschner et al., 2019). The final refinement converged with R_work of 20.42% and R_free of 23.45%. High R values are common for electron diffraction data due to dynamical effects. The no-filled 2F_o − F_c electrostatic potential map for the lysozyme structure is presented in Fig. 3(d).

Figure 7 Real-time 3D ED/microED data processing of triclinic lysozyme datasets using AutoLEI. (a) The live processing results are displayed under Live Statistics. (b) CC1/2 and (c) completeness against resolution plots captured from the web-page-based report.

4. Conclusion and outlook

We developed an XDS-based pipeline with a GUI, AutoLEI, for offline and real-time batch 3D ED/microED data processing. The visualization interface enables data import, processing, selection and merging with only a few clicks. We demonstrated the applications of AutoLEI on four different samples with various processing strategies to show its simplicity and ease of use. For stable or high-symmetry crystals, AutoLEI enables rapid identification of promising single datasets for structure determination. For low symmetry or electron-sensitive materials, AutoLEI can assist in selecting and merging datasets, thereby generating a final HKL file with high completeness. In addition, real-time data processing provides direct feedback, allowing users to adjust data collection strategies `on the fly'.

Further developments are planned for AutoLEI, including incorporation of additional data processing engines, enabling comparative analyses and data integration. These developments will further simplify 3D ED/microED data processing and lower the entry barriers for researchers new to the field of 3D ED/microED.

5. Related literature

The following references are cited in the supporting information: Burnley et al. (2017), Dolomanov et al. (2009), Giordano et al. (2012), Grant & Pickup (1995), Grosse-Kunstleve (1999), Grosse-Kunstleve et al. (2004), Jain & Dubes (1988), Knudsen et al. (2013), Kolb et al. (2012), Niggli (1928), Palatinus (2013), Schönemann (1966), Sheldrick (2008), Storn & Price (1997), Van Der Walt et al. (2014).

Supporting information

Link https://doi.org/10.5281/zenodo.14536385
Raw data used for examples

Link https://doi.org/10.5281/zenodo.15155987
AutoLEI software

Computing details

(SU-100)

Crystal data

C15H7BiO6	F(000) = 530
Mr = 492.19	Dx = 1.706 Mg m^−3
Monoclinic, C2/c	Electron radiation, λ = 0.02508 Å
a = 26.44 (3) Å	Cell parameters from 24457 reflections
b = 10.049 (17) Å	θ = 0.04–1.2°
c = 18.135 (17) Å	µ = 0.000 mm^−1
β = 127.28 (9)°	T = 100 K
V = 3833 (9) Å^3	Nano-crystal
Z = 8

Data collection

JEOL-2100 diffractometer	4231 reflections with I > 2σ(I)
Radiation source: electron microscope	Rint = 0.357
Graphite monochromator	θmax = 1.2°, θmin = 0.1°
Detector resolution: 18.1818 pixels mm^-1	h = −42→42
Continuous Rotation Electron Diffraction scans	k = −14→15
95728 measured reflections	l = −28→26
7193 independent reflections

Refinement

Refinement on F^2	Hydrogen site location: inferred from neighbouring sites
Least-squares matrix: full	H-atom parameters constrained
R[F^2 > 2σ(F^2)] = 0.195	w = 1/[σ^2(Fo^2) + (0.1653P)^2 + 7.8881P] where P = (Fo^2 + 2Fc^2)/3
wR(F^2) = 0.487	(Δ/σ)max < 0.001
S = 1.10	Δρmax = 0.56 e Å^−3
7193 reflections	Δρmin = −0.33 e Å^−3
200 parameters	Extinction correction: SHELXL-2018/3 (Sheldrick 2018), Fc^*=kFc[1+0.001xFc^2λ^3/sin(2θ)]^-1/4
0 restraints	Extinction coefficient: 637 (55)

Special details

Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.

Fractional atomic coordinates and isotropic or equivalent isotropic displacement parameters (Ų)

	x	y	z	Uiso*/Ueq
Bi2	−0.14106 (6)	1.18664 (18)	0.60987 (9)	0.0328 (5)
O1	0.2375 (4)	1.6974 (13)	0.5625 (6)	0.041 (3)
O3	−0.1294 (4)	0.8850 (11)	0.2536 (6)	0.036 (2)
C4	−0.0473 (4)	1.1526 (12)	0.4836 (7)	0.030 (2)
O5	−0.0773 (3)	0.9791 (10)	0.2057 (5)	0.031 (2)
C8	0.0820 (5)	1.3205 (14)	0.4744 (8)	0.037 (3)
C6	−0.0047 (4)	1.1405 (14)	0.3713 (7)	0.040 (3)
H6	0.012355	1.132806	0.330113	0.048*
O7	−0.0401 (4)	1.2293 (11)	0.6156 (6)	0.035 (2)
C9	0.0270 (5)	1.2272 (13)	0.4502 (7)	0.033 (3)
O10	0.2872 (3)	1.5217 (11)	0.5509 (6)	0.035 (2)
C11	−0.0805 (4)	1.0683 (13)	0.4006 (6)	0.034 (3)
H11	−0.121463	1.009982	0.380847	0.041*
C12	−0.0703 (4)	1.1604 (12)	0.5425 (7)	0.030 (3)
O15	−0.1217 (4)	1.0944 (11)	0.5156 (6)	0.036 (2)
C16	0.0914 (5)	1.4420 (14)	0.5162 (9)	0.044 (3)
H16	0.059135	1.476108	0.530162	0.053*
C17	−0.0581 (4)	1.0646 (13)	0.3468 (6)	0.033 (3)
C18	0.2388 (4)	1.5731 (14)	0.5460 (8)	0.038 (3)
C19	−0.0908 (4)	0.9715 (12)	0.2649 (7)	0.030 (2)
C20	0.1429 (5)	1.5208 (14)	0.5404 (9)	0.042 (3)
H20	0.151052	1.614498	0.575528	0.051*
C21	0.0045 (4)	1.2330 (13)	0.5068 (7)	0.033 (3)
H21	0.027378	1.298714	0.566257	0.040*
C3	0.1242 (5)	1.2748 (15)	0.4557 (8)	0.042 (3)
H3	0.116742	1.178328	0.423912	0.050*
C2	0.1762 (4)	1.3559 (13)	0.4786 (8)	0.038 (3)
H2	0.208487	1.321878	0.464579	0.046*
C5	0.1850 (4)	1.4811 (12)	0.5199 (7)	0.032 (3)

Atomic displacement parameters (Ų)

	U^11	U^22	U^33	U^12	U^13	U^23
Bi2	0.0243 (6)	0.0537 (11)	0.0304 (7)	−0.0023 (6)	0.0217 (5)	−0.0019 (7)
O1	0.029 (4)	0.062 (9)	0.037 (4)	−0.005 (4)	0.023 (3)	−0.001 (5)
O3	0.039 (4)	0.047 (7)	0.039 (4)	−0.003 (4)	0.032 (4)	0.005 (4)
C4	0.022 (4)	0.041 (8)	0.033 (5)	−0.009 (4)	0.020 (4)	−0.007 (5)
O5	0.023 (3)	0.051 (6)	0.029 (3)	0.000 (3)	0.020 (3)	−0.001 (4)
C8	0.028 (4)	0.044 (8)	0.041 (5)	0.005 (5)	0.023 (4)	−0.005 (6)
C6	0.028 (4)	0.066 (10)	0.033 (5)	−0.013 (5)	0.022 (4)	0.005 (5)
O7	0.029 (3)	0.051 (7)	0.031 (4)	−0.011 (4)	0.022 (3)	−0.007 (4)
C9	0.034 (5)	0.041 (8)	0.044 (6)	0.000 (4)	0.033 (5)	−0.001 (5)
O10	0.016 (3)	0.055 (7)	0.037 (4)	0.003 (3)	0.017 (3)	0.001 (4)
C11	0.026 (4)	0.054 (9)	0.031 (5)	−0.010 (4)	0.021 (4)	−0.002 (5)
C12	0.025 (4)	0.040 (8)	0.034 (5)	0.004 (4)	0.022 (4)	0.008 (5)
O15	0.031 (3)	0.056 (7)	0.036 (4)	−0.005 (4)	0.028 (3)	−0.004 (4)
C16	0.044 (6)	0.049 (9)	0.072 (8)	−0.011 (5)	0.053 (6)	−0.004 (7)
C17	0.016 (4)	0.056 (9)	0.027 (4)	0.002 (4)	0.013 (3)	0.000 (5)
C18	0.021 (4)	0.053 (9)	0.048 (6)	−0.001 (4)	0.026 (4)	0.007 (6)
C19	0.030 (4)	0.037 (8)	0.032 (5)	−0.004 (4)	0.023 (4)	0.003 (5)
C20	0.037 (5)	0.046 (9)	0.066 (7)	−0.008 (5)	0.043 (6)	−0.006 (6)
C21	0.025 (4)	0.048 (8)	0.032 (5)	−0.008 (4)	0.020 (4)	−0.001 (5)
C3	0.026 (4)	0.060 (10)	0.055 (7)	0.000 (5)	0.032 (5)	0.001 (6)
C2	0.025 (4)	0.047 (8)	0.054 (6)	−0.004 (4)	0.031 (5)	−0.009 (6)
C5	0.028 (4)	0.039 (8)	0.046 (5)	−0.004 (4)	0.031 (4)	−0.003 (5)

Geometric parameters (Å, º)

Bi2—O3^i	2.536 (9)	C9—C21	1.468 (13)
Bi2—O5^i	2.255 (10)	O10—C18	1.333 (12)
Bi2—O10^ii	2.242 (10)	C11—H11	1.0830
Bi2—O15	2.259 (8)	C11—C17	1.417 (12)
O1—C18	1.290 (17)	C12—O15	1.312 (13)
O3—C19	1.258 (14)	C16—H16	1.0830
C4—C11	1.467 (15)	C16—C20	1.395 (15)
C4—C12	1.517 (13)	C17—C19	1.507 (16)
C4—C21	1.416 (14)	C18—C5	1.512 (15)
O5—C19	1.324 (11)	C20—H20	1.0830
C8—C9	1.553 (16)	C20—C5	1.426 (13)
C8—C16	1.378 (18)	C21—H21	1.0830
C8—C3	1.424 (14)	C3—H3	1.0830
C6—H6	1.0830	C3—C2	1.427 (15)
C6—C9	1.435 (17)	C2—H2	1.0830
C6—C17	1.419 (14)	C2—C5	1.410 (17)
O7—C12	1.262 (14)
O5^i—Bi2—O3^i	55.2 (3)	C8—C16—H16	120.2
O5^i—Bi2—O15	82.7 (3)	C8—C16—C20	119.5 (10)
O10^ii—Bi2—O3^i	79.7 (3)	C20—C16—H16	120.2
O10^ii—Bi2—O5^i	81.0 (4)	C6—C17—C19	120.7 (8)
O10^ii—Bi2—O15	81.4 (3)	C11—C17—C6	121.5 (10)
O15—Bi2—O3^i	136.0 (4)	C11—C17—C19	117.7 (9)
C19—O3—Bi2^iii	86.9 (6)	O1—C18—O10	121.3 (10)
C11—C4—C12	120.4 (8)	O1—C18—C5	120.8 (9)
C21—C4—C11	120.3 (8)	O10—C18—C5	118.0 (12)
C21—C4—C12	119.2 (9)	O3—C19—O5	119.8 (10)
C19—O5—Bi2^iii	98.0 (7)	O3—C19—C17	121.0 (8)
C16—C8—C9	121.1 (8)	O5—C19—C17	119.2 (9)
C16—C8—C3	120.7 (11)	C16—C20—H20	119.2
C3—C8—C9	118.2 (11)	C16—C20—C5	121.7 (12)
C9—C6—H6	119.7	C5—C20—H20	119.2
C17—C6—H6	119.7	C4—C21—C9	119.9 (10)
C17—C6—C9	120.5 (9)	C4—C21—H21	120.1
C6—C9—C8	121.9 (8)	C9—C21—H21	120.1
C6—C9—C21	118.9 (9)	C8—C3—H3	119.9
C21—C9—C8	119.1 (10)	C8—C3—C2	120.1 (13)
C18—O10—Bi2^iv	104.3 (8)	C2—C3—H3	119.9
C4—C11—H11	120.6	C3—C2—H2	120.5
C17—C11—C4	118.8 (8)	C5—C2—C3	118.9 (9)
C17—C11—H11	120.6	C5—C2—H2	120.5
O7—C12—C4	120.1 (8)	C20—C5—C18	118.9 (11)
O7—C12—O15	121.3 (8)	C2—C5—C18	122.1 (8)
O15—C12—C4	118.6 (9)	C2—C5—C20	119.0 (9)
C12—O15—Bi2	99.8 (7)
Bi2^iii—O3—C19—O5	−0.8 (10)	O10—C18—C5—C2	−16.5 (16)
Bi2^iii—O3—C19—C17	−178.9 (10)	C11—C4—C12—O7	177.4 (11)
Bi2^iii—O5—C19—O3	0.9 (12)	C11—C4—C12—O15	−2.5 (16)
Bi2^iii—O5—C19—C17	179.0 (8)	C11—C4—C21—C9	−3.0 (17)
Bi2^iv—O10—C18—O1	7.8 (12)	C11—C17—C19—O3	−11.4 (16)
Bi2^iv—O10—C18—C5	−172.4 (8)	C11—C17—C19—O5	170.5 (10)
O1—C18—C5—C20	−18.7 (16)	C12—C4—C11—C17	179.3 (10)
O1—C18—C5—C2	163.2 (11)	C12—C4—C21—C9	−179.6 (10)
C4—C11—C17—C6	−0.2 (17)	C16—C8—C9—C6	148.0 (12)
C4—C11—C17—C19	176.7 (10)	C16—C8—C9—C21	−28.7 (17)
C4—C12—O15—Bi2	−167.1 (8)	C16—C8—C3—C2	−1.1 (18)
C8—C9—C21—C4	177.5 (11)	C16—C20—C5—C18	178.5 (11)
C8—C16—C20—C5	2.3 (19)	C16—C20—C5—C2	−3.4 (18)
C8—C3—C2—C5	0.0 (17)	C17—C6—C9—C8	−174.8 (11)
C6—C9—C21—C4	0.7 (17)	C17—C6—C9—C21	1.9 (18)
C6—C17—C19—O3	165.5 (11)	C21—C4—C11—C17	2.7 (17)
C6—C17—C19—O5	−12.6 (16)	C21—C4—C12—O7	−6.0 (16)
O7—C12—O15—Bi2	12.9 (13)	C21—C4—C12—O15	174.1 (11)
C9—C8—C16—C20	177.4 (11)	C3—C8—C9—C6	−34.6 (17)
C9—C8—C3—C2	−178.6 (10)	C3—C8—C9—C21	148.7 (11)
C9—C6—C17—C11	−2.2 (18)	C3—C8—C16—C20	0.0 (18)
C9—C6—C17—C19	−179.0 (11)	C3—C2—C5—C18	−179.7 (10)
O10—C18—C5—C20	161.6 (10)	C3—C2—C5—C20	2.1 (17)

Symmetry codes: (i) x, −y+2, z+1/2; (ii) x−1/2, y−1/2, z; (iii) x, −y+2, z−1/2; (iv) x+1/2, y+1/2, z.

(TyrosineExp10)

Crystal data

C9H11NO3	F(000) = 146
Mr = 181.19	Dx = 1.315 Mg m^−3
Orthorhombic, P212121	Electron radiation, λ = 0.02508 Å
a = 6.0080 (12) Å	Cell parameters from 1865 reflections
b = 7.0250 (14) Å	θ = 0.03–1.4°
c = 21.690 (4) Å	µ = 0.000 mm^−1
V = 915.5 (3) Å^3	T = 100 K
Z = 4	Nano-crystal

Data collection

TEM diffractometer	1253 reflections with I > 2σ(I)
Radiation source: electron microscope	Rint = 0.156
Graphite monochromator	θmax = 1.1°, θmin = 0.1°
Detector resolution: 18.1818 pixels mm^-1	h = −8→7
Continuous Rotation Electron Diffraction scans	k = −9→9
5506 measured reflections	l = −27→30
2592 independent reflections

Refinement

Refinement on F^2	Hydrogen site location: difference Fourier map
Least-squares matrix: full	All H-atom parameters refined
R[F^2 > 2σ(F^2)] = 0.119	w = 1/[σ^2(Fo^2) + (0.1456P)^2] where P = (Fo^2 + 2Fc^2)/3
wR(F^2) = 0.319	(Δ/σ)max < 0.001
S = 1.00	Δρmax = 0.15 e Å^−3
2592 reflections	Δρmin = −0.22 e Å^−3
162 parameters	Absolute structure: All f" are zero, so absolute structure could not be determined
8 restraints

Special details

Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.

Fractional atomic coordinates and isotropic or equivalent isotropic displacement parameters (Ų)

	x	y	z	Uiso*/Ueq
O24	0.4709 (9)	0.0196 (11)	0.0986 (3)	0.0175 (14)
C15	0.5630 (11)	−0.0964 (14)	−0.0669 (3)	0.0170 (17)
O18	0.7702 (9)	0.0750 (12)	−0.2137 (3)	0.0217 (16)
C1	0.3115 (11)	−0.0743 (14)	−0.1643 (3)	0.0170 (17)
N10	0.1954 (10)	0.2603 (13)	−0.1962 (3)	0.0195 (17)
C4	0.3557 (10)	−0.0408 (13)	−0.0944 (3)	0.0150 (17)
C20	0.4305 (10)	0.0050 (13)	0.0359 (3)	0.0163 (18)
C22	0.6001 (11)	−0.0735 (12)	−0.0025 (3)	0.0119 (15)
C14	0.1871 (10)	0.0441 (14)	−0.0551 (3)	0.0157 (18)
C16	0.6063 (11)	0.1780 (13)	−0.1942 (3)	0.0158 (17)
C23	0.3660 (10)	0.1040 (13)	−0.2064 (3)	0.0147 (17)
C12	0.2253 (11)	0.0673 (14)	0.0094 (3)	0.0175 (18)
O13	0.6238 (10)	0.3389 (12)	−0.1660 (3)	0.0212 (15)
H10A	0.210 (6)	0.327 (6)	−0.1535 (12)	0.017 (7)*
H23	0.348 (5)	0.060 (6)	−0.2548 (11)	0.018 (7)*
H14	0.025 (5)	0.072 (8)	−0.0772 (19)	0.031 (9)*
H22	0.753 (6)	−0.124 (9)	0.018 (2)	0.052 (14)*
H1A	0.410 (8)	−0.195 (8)	−0.180 (2)	0.034 (10)*
H1B	0.140 (5)	−0.123 (9)	−0.167 (3)	0.054 (14)*
H24	0.337 (9)	0.065 (11)	0.122 (3)	0.053 (14)*
H12	0.117 (12)	0.139 (13)	0.040 (3)	0.09 (2)*
H10B	0.214 (9)	0.362 (7)	−0.230 (2)	0.050 (14)*
H15	0.704 (6)	−0.145 (7)	−0.0955 (18)	0.026 (9)*
H10C	0.036 (7)	0.207 (11)	−0.203 (3)	0.067 (18)*

Atomic displacement parameters (Ų)

	U^11	U^22	U^33	U^12	U^13	U^23
O24	0.019 (2)	0.017 (5)	0.016 (3)	−0.001 (2)	0.000 (2)	−0.002 (2)
C15	0.014 (3)	0.019 (5)	0.018 (4)	−0.001 (3)	0.002 (3)	0.006 (3)
O18	0.013 (2)	0.031 (5)	0.021 (3)	0.004 (3)	0.0015 (19)	0.003 (3)
C1	0.016 (3)	0.023 (6)	0.012 (4)	−0.005 (3)	−0.001 (2)	0.005 (3)
N10	0.011 (2)	0.029 (5)	0.018 (4)	−0.001 (3)	0.000 (2)	0.003 (3)
C4	0.011 (3)	0.023 (6)	0.012 (3)	0.002 (3)	0.000 (2)	0.003 (3)
C20	0.011 (3)	0.022 (6)	0.016 (4)	−0.002 (3)	0.000 (3)	0.001 (3)
C22	0.012 (3)	0.008 (5)	0.016 (4)	−0.001 (3)	−0.003 (2)	0.002 (3)
C14	0.010 (3)	0.021 (6)	0.016 (4)	0.003 (3)	−0.002 (2)	0.007 (3)
C16	0.011 (3)	0.014 (5)	0.022 (4)	0.002 (3)	0.000 (3)	−0.005 (3)
C23	0.010 (3)	0.021 (5)	0.014 (4)	0.002 (3)	0.002 (2)	0.001 (3)
C12	0.013 (3)	0.027 (6)	0.012 (4)	0.000 (3)	0.001 (2)	0.002 (3)
O13	0.016 (2)	0.021 (4)	0.026 (4)	−0.003 (3)	−0.003 (2)	0.002 (3)

Geometric parameters (Å, º)

O24—C20	1.385 (10)	N10—H10B	1.02 (2)
O24—H24	1.01 (6)	N10—H10C	1.04 (3)
C15—C4	1.435 (9)	C4—C14	1.451 (10)
C15—C22	1.424 (10)	C20—C22	1.427 (10)
C15—H15	1.10 (4)	C20—C12	1.429 (10)
O18—C16	1.292 (10)	C22—H22	1.08 (2)
C1—C4	1.557 (10)	C14—C12	1.426 (10)
C1—C23	1.584 (12)	C14—H14	1.11 (2)
C1—H1A	1.09 (5)	C16—C23	1.557 (9)
C1—H1B	1.09 (2)	C16—O13	1.290 (12)
N10—C23	1.518 (11)	C23—H23	1.10 (2)
N10—H10A	1.04 (2)	C12—H12	1.06 (3)
C20—O24—H24	113 (3)	O24—C20—C12	121.6 (6)
C4—C15—H15	121 (2)	C22—C20—C12	119.9 (6)
C22—C15—C4	120.8 (7)	C15—C22—C20	120.3 (6)
C22—C15—H15	118 (2)	C15—C22—H22	120 (3)
C4—C1—C23	114.0 (7)	C20—C22—H22	120 (3)
C4—C1—H1A	109 (2)	C4—C14—H14	116 (2)
C4—C1—H1B	105 (3)	C12—C14—C4	120.6 (6)
C23—C1—H1A	109 (3)	C12—C14—H14	123 (2)
C23—C1—H1B	115 (4)	O18—C16—C23	117.7 (7)
H1A—C1—H1B	105 (5)	O13—C16—O18	125.6 (7)
C23—N10—H10A	113 (2)	O13—C16—C23	116.6 (6)
C23—N10—H10B	109 (3)	C1—C23—H23	108 (2)
C23—N10—H10C	110 (5)	N10—C23—C1	110.4 (6)
H10A—N10—H10B	108 (5)	N10—C23—C16	111.1 (7)
H10A—N10—H10C	112 (4)	N10—C23—H23	106 (2)
H10B—N10—H10C	105 (5)	C16—C23—C1	111.0 (6)
C15—C4—C1	120.7 (6)	C16—C23—H23	110.2 (17)
C15—C4—C14	118.3 (6)	C20—C12—H12	115 (5)
C14—C4—C1	121.0 (6)	C14—C12—C20	119.9 (7)
O24—C20—C22	118.5 (6)	C14—C12—H12	125 (5)
O24—C20—C22—C15	178.1 (8)	C4—C14—C12—C20	−0.5 (13)
O24—C20—C12—C14	−178.0 (8)	C22—C15—C4—C1	−177.2 (8)
C15—C4—C14—C12	−1.5 (13)	C22—C15—C4—C14	1.6 (13)
O18—C16—C23—C1	71.4 (9)	C22—C20—C12—C14	2.4 (13)
O18—C16—C23—N10	−165.3 (7)	C23—C1—C4—C15	−94.9 (9)
C1—C4—C14—C12	177.4 (8)	C23—C1—C4—C14	86.2 (9)
C4—C15—C22—C20	0.2 (13)	C12—C20—C22—C15	−2.2 (13)
C4—C1—C23—N10	−71.1 (7)	O13—C16—C23—C1	−109.2 (9)
C4—C1—C23—C16	52.6 (8)	O13—C16—C23—N10	14.1 (10)

 

Footnotes

‡These authors contributed equally to this work.