Automated and real-time structure solution using 3D electron diffraction

Three-dimensional electron diffraction (3D ED) has emerged as a powerful tool for solving the structures of small crystals down to nanometre-scale sizes. Despite advancements in automating data acquisition for 3D ED, the subsequent data processing and structure solution have largely relied on human intervention and have been mostly conducted offline. This reliance on expertise in electron crystallography and the lack of real-time feedback on data quality and structural information have limited the broader adoption of 3D ED. Here, we introduce Instamatic-solve, a fully automated, real-time structure solution pipeline for 3D ED deployed on a JEOL JEM 2100 transmission electron microscope. Instamatic-solve streamlines the entire process by automating the subsequent data processing and structure solution, providing real-time assessments of data quality and structural information. Moreover, the pipeline can handle offline 3D ED data acquired from various transmission electron microscope platforms. Using Instamatic-solve, we have successfully solved the crystal structures of diverse materials, including seven inorganic zeolites, two inorganic–organic hybrids and four organic molecules (including pharmaceuticals), all within 2 min. Instamatic-solve mimics the typical manual structure solution process, and its outcomes depend heavily on data quality. Our results indicate that a routine and reliable structure solution is achievable in most cases, provided that the data meet critical quality criteria, namely completeness ≥50% and resolution better than 1.0 Å. By enabling efficient, automated and real-time structure solution for crystalline materials, Instamatic-solve spans various scientific disciplines.

1. Introduction

For over a century, single-crystal X-ray diffraction (SCXRD) has been the primary technique for determining the structures of crystalline materials (Thomas, 2012; Gerstner, 2011). Modern SCXRD structure solution is routine and straightforward, aided by commercial and open-source automated real-time structure solution pipelines (Holton & Alber, 2004; Matsumoto et al., 2021; Viswanathan et al., 2019; Rigaku, 2020; Bruker, 2016). However, SCXRD requires sufficiently large crystals (>5 × 5 × 5 µm) to harvest adequate diffraction signals (McCusker & Baerlocher, 2013; Yang et al., 2022). For smaller, submicrometre-sized crystals, powder X-ray diffraction (PXRD), which measures diffraction from millions of crystals at once, is often used. Yet, the resulting peak overlaps can complicate structure solution, especially for those structures with low symmetries and/or large unit cells (McCusker & Baerlocher, 2013; Yang et al., 2022; Li & Sun, 2017; Zhou et al., 2016).

By contrast, the stronger interaction between electrons and matter makes electron diffraction well suited for structural studies of crystals that are too small for SCXRD (Gemmi et al., 2019). Three-dimensional electron diffraction (3D ED), which is conceptually similar to SCXRD, has been extensively used for structural studies in the past two decades. Several automated or semi-automated data collection protocols including automated diffraction tomography (Kolb et al., 2007), rotation electron diffraction (Zhang et al., 2010; Wan et al., 2013), continuous-rotation electron diffraction (cRED) (Wang, Yang et al., 2018; Cichocka et al., 2018), microcrystal electron diffraction (Nannenga et al., 2014; Nannenga, 2020), precession electron diffraction tomography (Gemmi et al., 2013; Kolb et al., 2019), serial electron diffraction (SerialED) (Smeets, 2018; Bücker et al., 2020; Hogan-Lamarre et al., 2024) and serial rotation electron diffraction (SerialRED) (Wang et al., 2019; Luo et al., 2023) have been developed for 3D ED. These methods have enabled successful structure determination of a wide range of polycrystalline materials [e.g. zeolites, metal–organic frameworks (MOFs) and drug molecules] which prove challenging for SCXRD and/or PXRD (Gemmi et al., 2019; Huang, Grape et al., 2021; Palatinus et al., 2017; Luo et al., 2017; Xu et al., 2019; Yonekura et al., 2015).

3D ED structural studies include the steps of (1) crystal screening, (2) data acquisition, (3) data processing and (4) structure solution, all of which require effort and expertise (Li & Sun, 2017; Gruene & Mugnaioli, 2021; Huang, Willhammar & Zou, 2021; Saha et al., 2022; Yonekura et al., 2023; Bengtsson et al., 2022). While efforts have been made to automate these steps and streamline the workflow to rival SCXRD in convenience and routine, current advancements primarily address crystal screening and data acquisition (Winter, 2010; Vonrhein et al., 2011). For instance, Instamatic, an open-source Python package, integrates the SerialRED and SerialED methods for automated crystal screening and electron diffraction data collection (Smeets, 2018; Wang et al., 2019; Luo et al., 2023). While real-time data processing is available for SerialRED, it has mainly been used for unit-cell determination. Further data analysis and dataset merging still rely on offline, stepwise and command-line-based operations. Consequently, data processing and structure solution in 3D ED remain largely manual and time consuming. Moreover, the absence of real-time automation tools for seamless 3D ED data processing and structure solution after data collection can lead to redundant or insufficient datasets, poor data quality and insufficient data completeness. These pose challenges for effective data collection and structural analysis using 3D ED and hamper its widespread adoption (Wang et al., 2019; Luo et al., 2023; Ito et al., 2021, Bengtsson et al., 2022).

Autochem is the first reported pipeline to integrate all four steps for a real-time and automated 3D ED structure solution. However, it relies on the commercial CrysAlis Pro software (Rigaku, 2020) and is currently limited to the XtaLAB Synergy-ED diffractometer. An open-source pipeline, AutoMicroED, has been proposed for semi-automated processing of offline 3D ED data, specifically for small-molecule structure solution (Powell et al., 2021). However, there is a strong need for a more accessible 3D ED pipeline that is capable of fully automated real-time data processing and structure solution.

To this end, here we introduce the structure solution pipeline Instamatic-solve. Instamatic-solve is built by interfacing calls to the XDS and SHELXT programs within Instamatic (Smeets et al., 2018; Sheldrick, 2015; Kabsch, 2010; Smeets, 2018). We validated its automated real-time structure solution performance using two zeolites and tested its automated offline structure solution capabilities on five zeolites, two inorganic–organic hybrids and four small molecules. The offline 3D ED data of these samples were collected on either JEOL or Thermo Fisher Scientific (TFS) transmission electron microscopes (TEMs) equipped with different detectors (ASI Timepix, Ceta D, OneView and TemCam XF416) (Gallagher-Jones et al., 2020; Luo, Clabbers et al., 2022; Gorelik et al., 2023). Using Instamatic-solve, the crystal structures of all these tested samples have been successfully solved within 2 min. Our results demonstrate that Instamatic-solve can offer correct structure solutions for data with high enough completeness (≥50%) and high resolution (better than 1.0 Å). When the data completeness is high (≥90%), the requirement on resolution can be reduced to 1.2 Å, which is in line with the requirements on data quality for SCXRD structure solution (Sheldrick, 1990). Instamatic-solve enables effective data processing and structure solution without any human interactions, facilitating the application of the technique by novices.

2. Results and discussion

2.1. Proposed automated structure solution pipeline Instamatic-solve

In the well established offline, stepwise and manual 3D ED structure solution workflow, XDS (Kabsch, 2010) is commonly used for data processing, while SHELXT (Sheldrick, 2015) is employed for structure solution, both requiring user expertise in crystallography [Fig. 1(a)]. In the previous development of SerialRED (Wang et al., 2019; Luo et al., 2023), XDS was linked to SerialRED within Instamatic for real-time data processing. However, the data completeness of most single 3D ED datasets obtained from SerialRED was insufficient for direct structure solution using SHELXT. In this context, real-time data processing primarily served for unit-cell determination, while further data analysis and dataset merging were required. To facilitate this, edtools was used to call XDS for offline, semi-automated data processing and phase analysis, which still required manual command-line input. Structure solution was then performed offline using SHELXT. To streamline the entire process to perform fully automated and real-time structure solution, we developed Instamatic-solve, by linking XDS and SHELXT to the cRED method, which is implemented in Instamatic for acquiring high-quality single 3D ED datasets [Fig. 1(b)]. Instamatic-solve is embedded within and largely based on Instamatic, and it also supports automated structure solution using offline 3D ED datasets collected from diverse transmission electron microscope setups.

Figure 1 Flowcharts illustrating the procedures of (a) a typical manual and offline structure solution process and (b) our proposed fully automated and real-time Instamatic-solve pipeline (implemented in Instamatic for the cRED method) for 3D ED.

For real-time automated structure solution, Instamatic-solve was deployed on a JEOL JEM 2100 TEM equipped with an ASI Timepix detector. For offline use, Instamatic-solve can be run on any computer running Windows 7 or later, provided the 3D ED data (in SMV, CBF, RAXIS, TIFF or other XDS-compatible formats) are available. Before each pipeline run, users must supply a preliminary chemical composition for SHELXT, while input of the space group and unit-cell parameters is optional (Fig. 2). The GUI of Instamatic with a provided chemical composition is shown in Fig. S1 (in the supporting information). During pipeline execution, the cRED method implemented in Instamatic is responsible for 3D ED data collection. It also automatically records all necessary experimental parameters and generates the required XDS.inp and SMV files without any user intervention for the real-time automated structure solution (Wang, Yang et al., 2018). For the offline automated workflow, users only need to provide a folder path of the existing 3D ED data files. Once the data are ready, either real time or offline, XDS automatically processes the data, determines or refines the unit-cell parameters (run only once), identifies the Laue group, and generates a standard SHELX.hkl intensity file. For the resolution cutoff, the default resolution range is defined from 20 to 0.8 Å in the XDS.inp file. Therefore, all datasets will initially be processed with a cutoff at 0.8 Å. Depending on the statistics for this resolution range shown in CORRECT.Lp, Instamatic-solve will adjust the resolution cutoff when generating the SHELX.hkl intensity files using XDSCONV.inp. If reflections at 0.8 Å do not meet the criteria of I/σ(I) ≥ 0.3 and CC_1/2 is larger than or equal to 0.5 and flagged with a star, the cutoff will be automatically reduced to the highest resolution that satisfies these criteria. Instamatic in the end generates SHELX.ins, enabling SHELXT to extract the required information and perform the structure solution, including atom-type and space-group assignments.

Figure 2 Flowchart depicting the real-time and offline automated structure solution pipeline Instamatic-solve for 3D ED. The automated data reduction steps are highlighted in light blue, and the structure solution steps are in orange.

Notably, Instamatic-solve mimics a typical manual workflow, using XDS for data processing and SHELXT for structure solution. As in SCXRD, the success of an automated 3D ED structure solution depends heavily on data quality and on the capabilities of XDS and SHELXT. For instance, if the data completeness is below 50%, XDS may incorrectly identify the unit-cell parameters or space groups. Likewise, if the resolution is lower than 1.2 Å, SHELXT may fail to deliver viable structure solutions (Sheldrick, 1990). Furthermore, for crystals with complex compositions, particularly those containing multiple elements or elements with similar atomic numbers, SHELXT may encounter challenges in assigning the correct atom types.

2.2. The reliability of Instamatic-solve

For real-time automated structure solution, Instamatic-solve was evaluated using two zeolites of known structure, SCM-25 (framework type -HOS) and faujasite (framework type FAU). SCM-25 is a germanosilicate, whereas FAU is an aluminosilicate. The real-time Instamatic-solve workflow is shown in Videos 1 and 2 in the supporting information. Because SCM-25 contains mixed Ge/Si sites and FAU contains mixed Al/Si sites, SHELXT cannot directly handle the mixed tetrahedral (T) sites. Both zeolites therefore were treated as pure silica ([SiO₂]_n), using [Si1O2] as the element composition for structure solution. SCM-25 was successfully solved in 5.0 min, encompassing data acquisition, data processing and structure solution. This time consumption is much shorter than the time typically required by standard SCXRD experiments (Bloch et al., 2015). The crystallographic data and structure solution results for SCM-25 are listed in Table 1, and the obtained framework is shown in Fig. 3. The 3D ED dataset exhibited 89.4% completeness (resolution 0.8 Å), and XDS automatically identified a C-centered orthorhombic cell (a = 14.65, b = 51.87, c = 13.10 Å). Solutions of the structures (SHELX .res files) with the same topology but different symmetries or settings (Cmmm, Cmm2, Amm2 and C222) were obtained. In the highest symmetry Cmmm, all ten framework T sites (including one disordered) and 24 framework O atoms in the asymmetric unit were directly located. Two O atoms associated with the disordered T site [defects, T(OT)₃OH species] were missing, and their positions can be inferred from established zeolite chemistry principles (Flanigen et al., 1991). These real-time structure solution results agree with earlier, manually obtained ones (Luo, Fu et al., 2022). For FAU, with data completeness of 99.2% and resolution of 0.8 Å, a cubic cell with a = 25.08 Å and space group Fd3m was identified by SHELXT. The complete framework structure, including one T atom and four O atoms in the asymmetric unit, was readily solved, and extra-framework species (water molecules and/or Na⁺ cations) were also located (Fig. 3).

Table 1
Summary of the crystallographic data and automated structure solution details for known zeolites

The 3D ED data of all the samples were collected using the cRED method implemented in Instamatic.

Sample	SCM-25	Zeolite Y	EMM-37	RUB-13	ZSM-5	PST-14	AlPO-34
Framework type code	-HOS	FAU	ETV	RTH	MFI	POR	CHA
Reported space group	Cmmm	Fd3m	P1	C2/m	Pnma	P421c	R3m
XDS-identified space group	C222	F432	P1	C2	P222	P422	R3
SHELXT-identified space group	Cmmm	Fd3m	P1	Cm	Pnma	P421c	R3
a (Å)	14.65 (6)	25.08 (15)	9.04 (2)	10.10 (1)	13.58 (4)	15.08 (1)	14.01 (2)
b (Å)	51.87 (12)	25.08 (15)	9.77 (5)	21.90 (1)	19.82 (7)	15.08 (1)	14.01 (2)
c (Å)	13.10 (8)	25.08 (15)	10.81 (1)	10.48	20.22 (7)	19.22 (5)	15.34 (1)
α (°)	90	90	104.37 (8)	90	90	90	90
β (°)	90	90	98.32 (19)	96.02 (1)	90	90	90
γ (°)	90	90	99.49 (4)	90	90	90	120
Composition	Si136O276	Si192O384	Si14O28	Si32O64	Si96O192	P32Al32O128	P18Al18O72
No. of atoms in asymmetric unit	34	7	21	25	38	24	6
Rotation range (°)	91.09	92.20	110.17	80.67	105.28	94.78	53.76
Running time (min)	0.92	0.55	0.67	0.58	0.50	0.53	0.30
Completeness (%)	89.4	99.2	55.8	47.7	87.2	94.7	60.3
No. of unique reflections	5077	820	2033	1149	5639	2489	716
Reflection-to-parameter ratio	42.7	51.2	23.9	13.5	37.8	47.0	28.6
I/σ(I)	3.64	6.30	7.11	2.63	2.57	4.47	6.24
Resolution (Å)	0.80	0.80	0.80	0.80	0.80	0.80	0.80
Rint (%)	12.9	24.7	6.1	27.7	14.2	17.8	8.0
Initial R1 (%, all data)†	30.8	25.9	35.7	25.7	30.0	33.3	28.2

†The R1 values were calculated from the initial structure models given by SHELXT, prior to refinement.

Figure 3 Observed electrostatic potential maps and corresponding framework structures of different zeolites solved using Instamatic-solve. Electrostatic potential maps were generated from the structure-factor data ( *.fcf) using the Fourier synthesis function in VESTA, revealing the 3D distribution of the electrostatic potential within the structure. SCM-25 (-HOS), zeolite Y (FAU), EMM-37 (ETV), RUB-13 (RTH), ZSM-5 (MFI), PST-14 (POR) and AlPO-34 (CHA) crystallize in orthorhombic, cubic, triclinic, monoclinic, orthorhombic, tetragonal and hexagonal systems, respectively. The electrostatic potential map for RTH is less well resolved, mainly due to the low data completeness (47.7%). The blue dashed frame highlights the disordered T atoms and missing O atoms in the SCM-25 framework. The red dashed frame indicates the positions of extra-framework water molecules and/or Na^+ cations in FAU.

To run the offline automated structure solution, the path to the folder containing the existing 3D ED data (SMV format) and the corresponding XDS.inp file should be provided to Instamatic. Its performance was assessed on the structure solution of five known zeolite structures with different crystal systems. These zeolites are of ETV, RTH, MFI, POR and CHA framework types, and their crystal systems are triclinic (P1), monoclinic (C2/m), orthorhombic (Pnma), tetragonal (P42₁c) and hexagonal (R3m), respectively (Kapaca et al., 2019; Seo et al., 2018; Baerlocher & McCusker, 2019). For Instamatic-solve input, ETV, RTH and MFI were treated as pure silica ([SiO₂]_n), whereas CHA and POR were considered aluminophosphates ([AlPO₄]_n). A demonstration of the offline automated structure solution is provided in Video 3 in the supporting information. The offline 3D ED datasets of these zeolites were all collected via the cRED method implemented in Instamatic and had data completeness of 55.8, 47.7, 87.2, 94.7 and 60.3%, respectively (at 0.8 Å resolution; Table 1). Despite some datasets approaching a low 50% completeness, all five framework structures were successfully solved within 1 min (Fig. 3 and Table 1). The space groups assigned by SHELXT for ETV, MFI and POR are consistent with the reported ones, while for RTH and CHA, subgroups (Cm and R3) were assigned instead of the previously published C2/m and R3m. Note that the symmetry of a zeolite framework may vary depending on the synthesis conditions and/or the presence of guest species (e.g. H₂O) within the pores (Hogan-Lamarre et al., 2024). The RTH and CHA samples used in this study may possess lower symmetries than previously reported, or their symmetries may have been reduced due to the removal of guest species under vacuum conditions in the TEM.

These results of the real-time and offline automated structure solution underscore the reliability and efficiency of Instamatic-solve for zeolite structures with varying data quality. The underlying principles of real-time and offline automated structure solution are comparable; their main distinction lies in their respective advantages and use cases. Real-time solutions provide immediate feedback on data quality and structural features during a data collection session, while offline solutions enable rapid structure determination once datasets are ready. Together, they offer a versatile platform for diverse research needs while minimizing manual intervention.

2.3. The capability and adaptability of Instamatic-solve

Encouraged by the successful automated structure solution results of inorganic zeolites, we further evaluated the capability of Instamatic-solve for the structure solution of an inorganic–organic hybrid MOF material CAU-36 and two organic small molecules (biotin and acetamino­phen) (Table 2). Their offline 3D ED datasets were all collected using the cRED method implemented in Instamatic. CAU-36 has a complex element composition of [Co1Ni1P1C1N1O1] (Wang, Rhauderwiek et al., 2018) and its data completeness is 94.9% (resolution 0.8 Å). SHELXT identified a tetragonal cell (a = b = 21.95, c = 8.74 Å) in space group P4c2. The complete framework structure, comprising 23 symmetry-independent atoms, was directly solved [Fig. 4(a), Video 4 in the supporting information]. However, due to the similar atomic numbers of Co and Ni, and of C, N and O, SHELXT failed to assign these atom types correctly in the structure. Nonetheless, the atom types can be assigned correctly and straightforwardly on the basis of the chemical information of the building blocks used in the synthesis. For acetamino­phen and biotin, Instamatic-solve was run with [C1N1O1] and [C1N1O1S1] as the respective inputs. With data completeness of around 90% (resolution 0.8 Å), all non-hydrogen atoms were located [Figs. 4(b) and 4(c)]. Similarly to the case of CAU-36, SHELXT could not distinguish between C, N and O atoms in either acetamino­phen or biotin, yet these can be unambiguously assigned by referring to their known molecular structures and the typical bond distances of C=O (1.20 Å), C—N (1.48 Å) and C—C (1.54 Å) (Welberry, 2021).

Table 2
Summary of the crystallographic data and automated structure solution details of CAU-36 and pharmaceuticals

The chemical compositions of CAU-36, acetamino­phen and biotin are difficult for SHELXT to assign correctly.

Sample	CAU-36	Acetamino­phen	Biotin
Reported space group	P4c2	P21/n	P212121
Identified space group (XDS)	C222	P2	P222
Identified space group (SHELXT)	P4c2	P21/n	P212121
a (Å)	21.95 (2)	7.11 (2)	5.30 (3)
b (Å)	21.95 (2)	9.38 (2)	10.47 (5)
c (Å)	8.74 (5)	11.72 (5)	21.34 (1)
α (°)	90	90.0	90.0
β (°)	90	97.19 (13)	90.0
γ (°)	90	90.0	90.0
Composition	C232N80O16P12	C32N4O8	C40N8O12S4
No. of atoms in the asymmetric unit	23	11	16
Running time (min)	0.40	0.38	0.28
Completeness (%)	94.9	91.1	89.5
No. of unique reflections	1886	1520	1307
Reflection-to-parameter ratio	21.4	33.8	20.1
I/σ(I)	4.23	3.05	3.56
Resolution (Å)	0.80	0.80	0.80
Rint (%)	24.6	14.3	13.8
Initial R1 (%)†	28.5	42.3	28.1

†The R1 values were calculated from the initial structure models given by SHELXT, prior to refinement.

Figure 4 Observed electrostatic potential maps and corresponding solved structures of (a) CAU-36, (b) acetamino­phen and (c) biotin.

The 3D ED datasets for the above zeolites, MOF and small molecules were all collected on a JEOL JEM 2100 TEM equipped with an ASI Timepix detector. In a TEM configuration for 3D ED data acquisition, the characteristics of the detector, such as its electron detection sensitivity, background levels, dynamical range and read-out speed, can affect the quality of the data and ultimately the results of the structure solution. To explore the adaptability of Instamatic-solve to other instrumentation, we tested additional offline 3D ED datasets obtained from different research groups using various TEM platforms. Specifically, SCM-34 (Luo, Clabbers et al., 2022) data were collected on a TFS Themis Z (OneView camera), PPEA {9,10-bis-[(perchlorophenyl)ethynyl]anthracene} (Gorelik et al., 2023) on a Glacios Cryo-TEM (Ceta-D camera) and AVAAGA peptide (Gallagher-Jones et al., 2020) on a Tecnai F30 (TemCam XF416 camera). By converting the raw data to SMV format and updating key parameters (detector distance, physical size of pixels, wavelength, rotation axis etc.) in the XDS.inp file, all structures have been successfully solved using Instamatic-solve.

SCM-34 is a hybrid aluminophosphate (|(C₆N₃H₁₃)₂|[P₄Al₂O₁₈H₆]) which crystallizes in the triclinic space group P1 and contains 21 symmetry-independent non-hydrogen atoms in its unit cell. The structure can be solved at 0.8 Å resolution with a data completeness of 51.1%, including all non-hydrogen atoms [Fig. S2(a)]. PPEA (C₃₀H₉Cl₁₀) has a monoclinic unit cell (space group P2₁/c) with 20 symmetry-independent non-hydrogen atoms and a data completeness of 68.1% (resolution 0.8 Å, Table 3). In the solved structure, all C and Cl positions were correctly assigned [Fig. S2(b)]. The AVAAGA peptide (C₁₀H₂₃N₃O₆) possesses a large ortho­rhombic unit cell (a = 11.36, b = 4.73, c = 39.59 Å, P2₁2₁2₁) (Gallagher-Jones et al., 2020). XDS assigned a monoclinic unit cell (a = 11.30, b = 4.70, c = 38.86 Å, β = 90.74, P2) owing to a low data completeness of 47.5% (resolution 0.8 Å). SHELXT determined the crystal structure with the correct molecule structure in the subgroup P2₁ [Fig. S2(c)].

Table 3
Summary of the crystallographic data and automated structure solution details for samples studied using different TEM platforms

The chemical compositions of SCM-34 and AVAAGA peptide are difficult for SHELXT to assign correctly.

Sample	9,10-bis-[(perchloro-phenyl)-ethynyl]-anthracene (PPEA)	SCM-34	AVAAGA peptide
Detector	Ceta D	OneView	TemCam XF416
Microscope	Glacios Cryo-TEM	FEI Themis Z	Tecnai F30
Reported space group	P21/c	P1	P212121
Identified space group (XDS)	P2	P1	P2
Identified space group (SHELXT)	P21/c	P1	P21
a (Å)	3.81 (1)	7.01 (2)	11.30 (3)
b (Å)	12.74 (1)	8.67 (1)	4.70 (5)
c (Å)	26.46 (1)	12.47 (1)	38.86 (10)
α (°)	90.0	101.91 (27)	90.0
β (°)	90.70 (10)	102.36 (14)	90.74 (10)
γ (°)	90.0	90.43 (24)	90.0
Composition	C15Cl5	C14N2Al3O18P5	C35O13N8
No. of atoms in asymmetric unit	20	22	56
Running time (min)	1.40	0.93	1.58
Completeness (%)	68.1	51.1	47.5
No. of unique reflections	1878	1513	2249
Reflection-to-parameter ratio	23.2	18.2	10.0
I/σ(I)	4.29	8.32	4.55
Resolution (Å)	0.80	0.80	0.80
Rint (%)	10.6	5.1	13.3
Initial R1 (%)†	25.7	27.9	32.5

†The R1 values were calculated from the initial structure models given by SHELXT, prior to refinement.

These results collectively highlight the robustness and adaptability of Instamatic-solve across a diverse range of TEM platforms and material types, demonstrating that the pipeline can accommodate variations in detector characteristics.

2.4. Criteria for a successful structure solution using Instamatic-solve

Instamatic-solve has demonstrated high reliability for both real-time and offline automated structure solution, accommodating various material types and adapting to different TEM platforms. Nonetheless, its performance is critically dependent on the quality of the 3D ED data. To assess this impact, we systematically tested Instamatic-solve with single, unmerged 3D ED datasets from ETV, RTH, MFI, FAU, POR, CHA, acetamino­phen and biotin at different resolution and completeness cutoffs (Tables S1 and S2).

These tests highlight three key factors, data completeness, data resolution and crystal symmetry, as crucial for successful automated structure solution. As demonstrated by our results, datasets with completeness ≥50% and resolution better than 1.0 Å show a high probability of yielding correct solutions (Tables 1, 2, 3, S1 and S2). These criteria are consistent with previous studies on the influence of data resolution and completeness on 3D ED structure solution (Ge et al., 2021). In lower-symmetry cases with limited data completeness, success is highly sensitive to the decrease in resolution (data completeness remains nearly constant). For example, ETV (P1 symmetry, 55.8% completeness) could not be solved when the resolution cutoff increased from 1.0 to 1.2 Å, and solution of RTH (Cm symmetry, 47.7% completeness) only succeeded at 0.8 Å. By contrast, successful solutions of higher-symmetry and/or higher-completeness cases such as MFI (Pnma, 87.2%), FAU (Fd3m, 99.2%), POR (P42₁c, 94.7%) and CHA (R3, 60.3%) were achieved even with the resolution cutoff raised to 1.2 Å. Moreover, for datasets at 0.8 Å resolution, reducing completeness from about 90 to 50% still allowed most structures to be correctly solved (Table S2).

Within a given unit-cell volume, higher-symmetry structures generally involve fewer parameters and thus impose less stringent requirements on both resolution (Table S1) and completeness (Table S2). For instance, POR (P42₁c) was correctly solved with just 40% completeness (resolution 0.8 Å) or a resolution of 1.3 Å (completeness 93.4%). Furthermore, the reduced angular coverage requirement of high-symmetry crystals allows more complete data collection within the limited rotation range of TEM stages (typically up to ∼120°). In contrast, low-symmetry systems (such as ETV and RTH) pose greater challenges in acquiring data with high completeness. An analysis of the ratio between unique reflections and parameters (N_reflections/N_parameters) suggests that values ≥10 typically correlate with correct structure solutions (Tables S1 and S2). The number of parameters refers to atomic coordinates (x, y, z) and isotropic displacement parameters (U_iso) that need to be determined in structure solution. Parameters for atoms located at special positions, where certain coordinate values are fixed by symmetry (e.g. 0.25, 0.5, 1.0 etc.), were excluded from the count. These criteria also apply to small molecules such as acetamino­phen and biotin, mirroring the trends observed in inorganic zeolites.

Nowadays, the quality of 3D ED data from many sub­micrometre-sized crystals can easily meet the level required for successful automated structure solution with Instamatic-solve. The dual-space method employed by SHELXT has considerably lowered the data quality threshold while increasing structure solution reliability. Moreover, Instamatic-solve offers real-time assessments of data quality during a data collection session by parsing data resolution and completeness from the CORRECT.Lp file output by XDS. This immediate feedback helps users determine when sufficient data have been collected, thereby improving the efficiency of data collection and the structure solution.

The source code of Instamatic-solve is publicly available, allowing users to insert additional command flags for more complex dataset analyses. By mimicking manual workflows through XDS for data processing and SHELXT for structure solution, Instamatic-solve delivers results comparable to manual methods yet offers significantly greater convenience and speed. Although it currently does not provide automated structure refinement, the solved structures are sufficiently accurate to provide key atomic-scale insights into crystalline materials.

3. Code availability

Instamatic-solve, which is implemented in the Instamatic software Python package, is available from https://github.com/Junschen1/instamatic under the terms of GNU General Public License v.3.0. For detailed installation and usage instructions, please refer to the Supplementary Notes in the supporting information.

4. Conclusions

In summary, we present Instamatic-solve, a fully automated, real-time structure solution pipeline for 3D ED structure solution. By integrating the programs XDS and SHELXT with Instamatic, a Python package for automated electron diffraction data collection, this pipeline streamlines the entire workflow from data collection and data processing to structure solution. Instamatic-solve has successfully and efficiently solved the structures of a wide range of crystalline materials, including inorganic zeolites, inorganic–organic hybrids, and organic pharmaceuticals and peptides, regardless of variations in crystal symmetry and data quality. Our systematic tests indicate that Instamatic-solve consistently delivers correct structure solutions when 3D ED data meet quality criteria of completeness ≥50% and resolution better than 1.0 Å. For high-symmetry structures, it is inherently easier to obtain data with a high completeness and high reflection-to-parameter ratio, enabling a higher success rate of structure solution. Instamatic-solve can be adapted to JEOL and TFS TEM platforms with different detectors (ASI Timepix, Ceta-D, OneView and TemCam XF416). Notably, even users with limited expertise in crystallography can efficiently perform structure solution and obtain rapid structural insights. We anticipate that Instamatic-solve will not only advance the use of 3D ED techniques but also open up new opportunities for crystalline materials development.

5. Related literature

The following references are cited in the supporting information: Karplus & Diederichs (2012), Roslova et al. (2020), Wang et al. (2025).

Footnotes

‡These authors contributed equally.