Probabilistic Alignment of Spatial Transcriptomic Experiments for MALDI Imaging Serial Section Alignment

Significance of the topic

The alignment and three-dimensional reconstruction of serial tissue sections acquired by MALDI imaging addresses a core challenge in spatial molecular analysis: interpreting how chemical distributions change across tissue depth rather than only across individual slices. Volumetric reconstruction improves biological interpretation of gradients, spatially-resolved metabolite and lipid distributions, and morphological correlation, enabling more accurate insight into tissue organization, pathology progression, and multimodal integration with other imaging modalities.

Goals and study overview

This work demonstrates an automated framework to align and reconstruct volumetric MALDI imaging data from serial tissue sections. The primary aims are to (1) integrate the PASTE probabilistic alignment algorithm with SCiLS Lab via the SCiLS REST API, (2) enable automated, robust slice-to-slice registration using both spatial coordinates and spectral intensity information, and (3) visualize volumetric distributions of selected analytes across multiple aligned slices without requiring changes to typical MALDI acquisition workflows.

Methods and used instrumentation

Sample preparation and acquisition:

• Mouse brain tissue sectioned at 10 µm and thaw-mounted on IntelliSlides®.
• Data acquired on a timsTOF fleX instrument equipped with MALDI-2, microGRID, and the smartbeam 3D laser; acquisition used 20 µm spatial resolution and an m/z range of 50–1,000.

Computational workflow and software:

• Initial data visualization, segmentation and extraction of region-level intensity features performed in SCiLS Lab using bisecting k-means segmentation on a panel of ~24 metabolite/lipid features.
• Segmented region coordinates and intensity information exported through the SCiLS Lab REST API for external processing.
• Rough alignment performed using spatial heuristic clustering on xy-coordinates to stabilize tissue structure between slices.
• Refined probabilistic alignment implemented using the PASTE algorithm (Probabilistic Alignment of Spatial Transcriptomics Experiments) leveraging both position and metabolite/lipid intensity to compute optimal transport models for slice registration.
• Final aligned slices were voxelized and volumetrically reconstructed and visualized in SCiLS Lab; analysis orchestrated via SCiLS API and Python.

Main results and discussion

Key outcomes reported in this study include:

• Successful automated alignment and volumetric reconstruction of 22 serial sections acquired at 15 pixels/s (timing reported for timsTOF fleX with microGRID and smartbeam 3D), demonstrating throughput compatible with routine MALDI imaging workflows.
• Visualization examples show voxelized 3D distributions of metabolites such as ascorbate, glutathione and lipid LPE (18:1), both individually and as multi-analyte overlays, illustrating spatial gradients through tissue depth.
• Quality assessment of alignment used incremental displacement metrics across slices; results show continuous displacement trends with larger discontinuities between early slides (e.g., Slide 1 to Slide 2), indicating where registration is most challenged.
• The two-stage strategy—spatial heuristic clustering followed by intensity-informed optimal transport (PASTE)—stabilizes large morphological features before applying intensity-driven fine alignment, improving robustness over purely coordinate- or purely intensity-based approaches.

Discussion points and practical observations:

• The approach requires no changes to existing MALDI acquisition protocols; previously acquired datasets can be reprocessed to obtain volumetric reconstructions.
• Using segmentation-derived regions reduces computational complexity by focusing alignment on morphologically distinct areas rather than on full-resolution raw images.
• Integration through SCiLS API allows streamlined data exchange between commercial visualization software and custom alignment code (Python + PASTE), enabling reproducible pipelines.
• Displacements and alignment diagnostics provide actionable feedback to identify slices with preparation artifacts or registration failure modes.

Benefits and practical applications

The presented framework offers several practical advantages:

• Enhanced biological insight: Volumetric maps reveal depth-dependent metabolite and lipid gradients that single-slice analyses can miss.
• Retrospective analysis: Existing MALDI datasets can be aligned and reconstructed without re-acquisition, extending value of archived data.
• Multimodal integration: The probabilistic alignment can be applied to integrate datasets from different instrument modalities (e.g., spatial transcriptomics, complementary imaging MS platforms), enabling richer molecular context.
• Automation and scalability: API-based integration and algorithmic registration support processing of tens of serial sections with minimal manual intervention, suitable for studies requiring 3D molecular phenotyping.

Future trends and applications

Potential developments and opportunities include:

• Deeper multimodal registration combining spatial transcriptomics, immunohistochemistry, and imaging MS to map molecular signatures across modalities in 3D.
• Improved robustness using machine-learning based feature extraction to augment segmentation and provide more discriminative anchors for alignment.
• Real-time or near-real-time alignment feedback during acquisition to guide sectioning or re-imaging of problematic regions.
• Standardized volumetric data formats and visualization tools to facilitate sharing and comparative studies across laboratories.
• Application to larger tissue volumes and clinical cohorts to study disease progression, tumor heterogeneity, and pharmacokinetic distributions in three dimensions.

Conclusion

This work demonstrates a practical and automated pipeline for probabilistic alignment and volumetric reconstruction of MALDI imaging serial sections by integrating the PASTE algorithm with SCiLS Lab through the SCiLS API. The method preserves standard acquisition workflows, enables retrospective 3D analysis of metabolites and lipids, and supports multimodal data integration. Quantitative diagnostics of displacement and the two-stage registration strategy improve alignment quality and make the approach broadly applicable to research and translational studies that require spatially resolved volumetric molecular information.

References

• PASTE (Probabilistic Alignment of Spatial Transcriptomics Experiments); Raphael group, Department of Computer Science, Princeton University. Method referenced as the algorithmic basis for intensity-informed probabilistic alignment.
• Bruker Scientific LLC. Application and methods for MALDI imaging serial section acquisition and volumetric reconstruction. © 2026 Bruker.