PaperlayerView exact revision on arXiv

Source-linked AI summary

MultiPASE: Shape-Aware Comparative 3D Genome Analysis

N. Nipu, G. Johnson, C. Roth, D. Rogers, S. R. Starkenburg, C. R. Steadman, G. E. Marai

arXiv:2610.22786v1q-bio.QMcs.HC

TL;DR

Comparing reconstructed 3D genome structures across conditions and time is challenging because of noise, variability, arbitrary orientation, and limits of contact-matrix summaries for preserving spatial geometry. MultiPASE aligns gene-level point clouds, derives multi-scale shape representations, and integrates them with expression and accessibility data. In a viral-infection usage case, it revealed transient chromosome 1 reorganization, while supplementary validation supported the shape abstractions and the need for alignment.

  • Problem

    Comparisons of reconstructed 3D genome structures are challenged by noise, variability, and arbitrary orientation, while contact-matrix summaries do not preserve spatial geometry or readily integrate functional measurements.

    Source: §1 · p. 1

  • Method

    MultiPASE aligns gene-level point clouds into a shared frame and extracts HDR and PF shape abstractions for comparative analysis integrated with expression and chromatin accessibility.

    Source: §1 · p. 2

  • Results

    In the usage case, MultiPASE revealed transient chromosome 1 reorganization at 18 hours followed by partial return at 24 hours, and supplementary validation found HDR and PF preserved non-convex structure and alignment necessary for meaningful comparison.

    Source: §5 · p. 7

  • Takeaways & Limitations

    MultiPASE supports multi-scale comparison across conditions and time points while connecting spatial organization with gene expression and chromatin accessibility.

    Source: §5 · p. 8

    The method inherits uncertainty from input reconstructions; rigid alignment may miss large non-rigid changes, smoothing can reduce fine detail, and 2D projections collapse depth.

    Source: §5 · pp. 7–8

Abstract

from arXiv · show

3D genome structures reconstructed from Hi-C data provide valuable insights into chromatin organization, but comparing them across conditions and time points remains challenging because of noise, variability, and arbitrary orientation. State-of-the-art comparison approaches often rely on contact-matrix-based summaries, which do not preserve the spatial geometry of the reconstructed structures and make integration with gene expression and chromatin accessibility difficult. There is therefore a need for robust structural alignment and spatial abstraction methods that support integrated analysis across conditions and time points. We present MultiPASE, a computational pipeline for comparative analysis of reconstructed 3D genome structures represented as gene-level spatial point clouds. MultiPASE aligns point clouds into a shared coordinate frame using coarse-to-fine rigid registration with PCA-based axis disambiguation and robust ICP refinement, then summarizes spatial organization using two complementary shape abstraction methods that produce stable, nested, multi-scale representations from dense core regions to broader structural extent. Using Hi-C-derived reconstructions from viral infection experiments, we show how the approach reveals condition-dependent structural changes that are difficult to interpret from raw 3D structures alone. By integrating spatial structure with gene expression and chromatin accessibility, MultiPASE supports interactive analysis of spatial organization and gene relocation patterns. A quantitative validation, reported in the Supplementary Materials, shows that the proposed shape abstractions preserve non-convex features, remain stable across samples, and that the alignment pipeline is necessary for meaningful comparison. A usage case and a usability evaluation with eight participants demonstrate the pipeline's effectiveness for comparative structural analysis.

1. Introduction

Comparing reconstructed 3D genome structures across conditions is difficult because of noise and experimental variability, while common summaries can lose spatial geometry and separate structure from functional measurements. MultiPASE aligns gene-level point clouds, derives shape-based representations, and supports integrated visual analysis across conditions and time points.

1. Introduction

  • Reconstructed structures are difficult to compare across conditions because of noise, experimental variability, and artifacts.Existing approaches often summarize Hi-C contact matrices rather than reconstructed spatial geometry, while visualization support for alignment and systematic comparison remains limited.Source: §1 · p. 1
  • Reconstructed 3D genome structures are noisy, unregistered point clouds with arbitrary orientation and variable point counts, requiring alignment, abstraction, and aggregation before comparison.The paper frames these structures as second-generation ensemble data, where representations must be derived through transformation for comparison.Source: §1 · p. 2
  • MultiPASE integrates gene-level expression and chromatin accessibility measurements with spatial structure and provides a web-based visual analytics frontend.The frontend supports exploration across conditions and time points.Source: §1 · p. 2
  • MultiPASE aligns reconstructed point clouds into a shared coordinate frame and provides HDR and PF shape abstractions for multi-scale structural comparison.Its pipeline uses PCA-based axis disambiguation and rigid ICP refinement, with quantitative comparison metrics.Source: §1 · p. 2
  • The authors evaluate MultiPASE through quantitative validation, a viral-infection usage case, and a usability study.The validation, usage case, and usability study are reported in the Supplementary Materials.Source: §1 · p. 2

2. Related Work

Prior genome visualization and comparison work commonly focuses on sequences, contact matrices, or individual structures, with limited support for shape-based comparison of reconstructed structures across conditions and time. MultiPASE addresses these needs through point-cloud alignment and nested, multi-scale shape abstraction.

Ensemble Visualization

  • Second-generation ensemble methods derive spatial representations through transformation and aggregation, a setting shared by noisy, unregistered genome point clouds with arbitrary orientation and variable point counts.MultiPASE addresses this setting through alignment and multi-scale shape abstraction.Source: §2 · p. 2

3D Genome Visualization and Comparison

  • Genome visualization systems commonly focus on sequences, Hi-C contact matrices, or individual reconstructed structures, and offer limited alignment and shape-based comparison across conditions and time.MultiPASE instead works directly on reconstructed gene-level point clouds and connects spatial comparison with expression and accessibility.Source: §2 · p. 2

Point Cloud Registration and Shape Analysis

  • MultiPASE uses PCA-based initialization and trimmed rigid ICP for alignment, separating registration from later HDR and PF smoothing to preserve structural differences.The paper avoids non-rigid registration because it could absorb condition-dependent structural differences.Source: §2 · p. 2
  • MultiPASE uses HDR and PF to derive nested, multi-scale shape representations rather than alpha shapes, whose scale parameter can yield fragmented or inflated boundaries across noisy point clouds.The related work also describes shape contexts, spin images, and point feature histograms as methods for local point-cloud geometry.Source: §2 · p. 2

3.1. Input Data and Preprocessing

The analysis uses Hi-C-derived 3D genome reconstructions from uninfected and vaccinia-virus-infected African green monkey cells at multiple time points. Genes are mapped to the reconstructed structures and paired with expression and accessibility measurements for integrated comparison.

3.1. Input Data and Preprocessing

  • The study analyzes Vero Hi-C reconstructions from uninfected and vaccinia-virus-infected African green monkey cells sampled at 12, 18, and 24 hours post-infection.The source study had two biological replicates per condition and time point; the analysis used one Hi-C replicate per condition-time combination and included 24 chromosomes with available reconstructions and matched measurements.Source: §3.1 · pp. 2–3
  • The 4DGB workflow and LAMMPS generated 100 kb bead-based chromosome reconstructions, with each bead representing a contiguous 100 kb genomic locus.The resulting chromosome-level 3D coordinates form the structural input for subsequent analysis.Source: §3.1 · p. 3
  • Genes were represented by interpolated midpoints between neighboring bead coordinates and integrated with expression differences and ATAC-seq accessibility measurements.Accessibility was assigned using the peak closest to each gene’s transcription start site; gene-position precision is limited by reconstruction resolution.Source: §3.1 · p. 3

3.2. Task Analysis

Domain-expert discussions identified six analytical tasks that shaped MultiPASE’s pipeline and visual interface. The tasks span structural alignment and comparison, temporal reorganization, regulatory behavior, and links between spatial changes and functional measurements.

3.2. Task Analysis

  • Six expert-informed tasks cover alignment, stable shape summaries, multi-scale structural differences, temporal reorganization, regulatory-behavior patterns, and links to expression and accessibility.The tasks were identified through discussions with domain experts in viral infection genomics.Source: §3.2 · p. 3
  • Tasks T1–T3 motivated the computational pipeline, while T4–T6 guided the coordinated visual interface..Source: §3.2 · p. 3
  • The usability evaluation tasks were derived from the analytical needs identified in the task analysis.Source: §3.2 · p. 3

3.3. Temporal Grouping and Agreement Classes

Genes are grouped by comparing expression and TSS-based accessibility directions over three time points, yielding seven temporal agreement classes for later spatial analysis.

3.3. Temporal Grouping and Agreement Classes

  • Seven classes—both increase, both decrease, expression-only, accessibility-only, mixed, conflict, and stable—label genes for subsequent analysis of their 3D organization.At each time point, expression and accessibility are labeled increased, decreased, or unchanged, then compared as changing together, oppositely, or independently.Source: §3.3 · p. 3

3.4. MultiPASE: Alignment and Shape-based Abstraction of 3D Genome Structures

MultiPASE transforms gene-level 3D point clouds from Hi-C-derived reconstructions into aligned, comparable shape abstractions and structural-difference metrics. It combines rigid registration with shared projections and complementary density-based representations.

3.4.1. Structural Alignment

  • Centering, PCA axis disambiguation, and trimmed rigid ICP align each chromosome to the earliest uninfected time point while preserving structural differences.PCA alignment tests proper rotations; ICP removes the 10% largest residuals to reduce noise and outlier influence.Source: §3.4.1 · p. 4

3.4.2. Shared Scaling and Grid Construction

  • After shared-scale normalization, structures use identical grid bins and XY, YZ, and XZ projections; these fixed views aid comparison but collapse depth.The projections complement rather than replace the 3D structures; the default grid has 160 bins along its longest axis.Source: §3.4.2 · p. 4, §3.4.3 · p. 4

3.4.3. Density-based Shape Abstraction

  • HDR averages Gaussian-smoothed bootstrap density maps, removes values below 0.2% of the maximum, and extracts nested regions at target density-mass levels.The default is 256 resamples; lower mass fractions show dense cores and higher fractions capture broader extent.Source: §3.4.3 · p. 4

3.4.4. Point-fraction Shape Abstraction

  • PF ranks genes by local point density and retains the top fN points, producing nested shapes from dense cores to broader structural extent.The retained points are rasterized and their masks refined before extracting the outer contour; the default kernel bandwidth is half the median pairwise distance from a subsample.Source: §3.4.4 · pp. 4–5

3.4.5. Quantitative Comparison Metrics

  • Matched shapes are compared separately by projection and abstraction level using IoU for overlap, Hausdorff distance for maximum boundary separation, and meanNN for typical separation.Levels range from 100% to 50%, spanning dense cores to broader structural extent.Source: §3.4.5 · p. 5

3.5. System Design and Implementation

The system pairs a computational backend with an interactive visual frontend for exploring aligned structures and their associated shape and comparison outputs.

3.5. System Design and Implementation

  • The backend processes structures and functional measurements, aligns them to the earliest uninfected time point, extracts shapes, and computes structural-difference metrics for frontend exploration.The backend and MultiPASE are implemented in Python; the frontend uses React, D3.js, and Three.js.Source: §3.5 · pp. 5–6

3.6. Data Visualization Design

The frontend coordinates structural, shape, and temporal views for genome exploration, with navigation across species, chromosomes, and genes and support for saving or exporting analysis sessions.

3.6. Data Visualization Design

  • The 3D structure panel compares infected and uninfected structures across time, with synchronized navigation, temporal-class filtering, and gene-level spatial context.Genes appear as timepoint-colored points along a chromatin tube; hovering over a gene links its context to the shape panel.Source: §3.6 · p. 6
  • The shape panel overlays HDR or PF summaries on background masks across XY, YZ, and XZ projections, with adjustable abstraction levels and temporal-class filtering.Users can view change across time points or conditions; hovering over a shape connects it to the corresponding 3D structure.Source: §3.6 · p. 6
  • The temporal panel pairs an expression-accessibility scatterplot with gene-level trends, enabling inspection of their relationship across genes and over time.Genes are colored by temporal class, and selecting a subset filters other panels.Source: §3.6 · p. 6, Figure 3 · p. 6

4. Evaluation and Results

The usage case compares chromosome 1 under uninfected and infected conditions at 12, 18, and 24 hours, finding the largest infection-associated structural deviation at 18 hours. Shape views capture changes in gene concentration, while temporal regulatory views show that expression and accessibility changes do not always align.

4. Evaluation and Results

  • Quantitative validation of alignment and shape abstraction, along with the usability study, is reported in the Supplementary Materials.Source: §4 · p. 6

4.1. Usage Case Demonstration

  • At 18 hours, infected YZ comparisons had IoU 0.502 and meanNN 5.24 from 12hrs to 18hrs, while uninfected comparisons had IoU 0.592 to 0.601 and meanNN 2.86 to 4.15.The demonstration used publicly available African green monkey data for chromosome 1 at 12, 18, and 24 hours.Source: §4.1 · p. 6
  • Infection was associated with transient chromosome 1 reorganization: the 3D view showed elongation at 18 hours followed by partial reversion at 24 hours, which shape abstractions also captured.Source: §4.1 · pp. 6–7
  • At 18 hours, the 60% HDR core gene region expanded and contracted by 24 hours; the 60% PF showed a similar pattern over a broader area.At 60%, IoU / meanNN were 0.240 / 5.07 in YZ and 0.179 / 22.70 in XZ; XZ meanNN peaked at 55.03 from 18 to 24 hours.Source: §4.1 · p. 7, Figure 4 · p. 7
  • Genes with large expression changes often showed minimal accessibility changes, while genes with higher accessibility changes tended to show smaller expression changes.Gene-level views also showed cases of large accessibility changes without corresponding expression changes, consistent with prior dataset analyses [V∗24].Source: §4.1 · p. 7

5. Discussion and Conclusion

MultiPASE combines rigid alignment, multi-scale shape abstraction, quantitative comparison, and coordinated visual analysis to compare reconstructed genome structures across conditions and time points. The usage case shows transient chromosome 1 reorganization and gene-concentration changes, while supplementary validation supports the abstractions' stability and practical use.

5. Discussion and Conclusion

  • In the usage case, MultiPASE revealed transient chromosome 1 reorganization at 18 hours followed by partial return at 24 hours, and shape views exposed concentration changes difficult to interpret directly in 3D.Temporal views also supported prior observations that large expression changes do not necessarily correspond to accessibility changes [V∗24].Source: §5 · p. 7
  • Supplementary evaluation found that HDR and PF preserve non-convex structure, alignment is necessary for meaningful comparison, and relationships remain consistent across abstraction levels.Together with the usability evaluation, these results support MultiPASE's stability and practical use.Source: §5 · p. 7
  • MultiPASE inherits reconstruction uncertainty; rigid alignment may miss large non-rigid changes, PCA initialization can be ambiguous for nearly symmetric clouds, and smoothing and 2D projection can reduce detail.Gene positions are limited by the reconstructed bead model's 100 kb resolution.Source: §5 · p. 7
  • The conclusion presents MultiPASE as a pipeline and visual analytics tool connecting spatial organization with gene expression and chromatin accessibility across conditions and time points.Future work will evaluate additional datasets and reconstruction pipelines and explore direct 3D shape abstraction and alternative rigid registration methods.Source: §5 · p. 8
Loading 2610.22786v1…