WHY MATHEMATICS?

Why Mathematics? | SOMDE for Scalable Spatial Genes: Self-Organizing Maps, Metaspot Compression, Gaussian Processes and Resolution Trade-Offs

How SOMDE compresses thousands of spatial measurements into topology-preserving nodes before testing spatially variable genes, and what detail can be lost for speed.

Why mathematics is important becomes easier to see when a real decision depends on it. This guide follows the mechanism from definitions and dimensions to a worked calculation, validation, practical learning and responsible use. It keeps factual evidence separate from interpretation and treats every illustrative scenario as an example rather than a promise of results.

Choose a reading route

Train a self-organizing map on physical coordinates
Condense nearby sites into adjustable-resolution nodes
Construct sparse-aware meta-expression per node
Fit Gaussian-process spatial variance on the condensed map

Chapter map

1. Spatial input sites
2. SOM grid
3. Node coordinate weights
4. Winner node
5. Neighbourhood update
6. Topology preservation
7. Site-to-node groups
8. Mean expression
9. Maximum expression
10. Meta-expression mixture
11. Adjustable resolution
12. Node Gaussian process
13. Length-scale models
14. Spatial and noise variance
15. Back-mapping patterns
16. Governance and learning

Visit the How Mathematics Works hub for the wider series and its connections to education, reasoning, science, computing, design and careers.

1. Spatial input sites

The central mathematical idea is concrete. Each cell or spot has coordinates and gene expression.

A concrete situation

Twenty thousand sites cover one section.

What the mathematics does not license

Coordinates may contain gaps or fields of view. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Plot the mask. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

2. SOM grid

Here the geometry becomes operational. A regular k-by-k node layout initializes the condensed map.

A concrete situation

A ten-by-ten grid creates one hundred nodes.

What the mathematics does not license

Grid topology imposes structure. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Compare shapes and k. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

3. Node coordinate weights

This step deserves a careful pause. Every node stores a learnable coordinate vector.

A concrete situation

Weights begin across the tissue range.

What the mathematics does not license

Initialization affects convergence. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Fix seeds and restarts. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

4. Winner node

The useful question is not only how to calculate, but what the calculation assumes. Each site maps to its closest coordinate weight.

A concrete situation

One site selects node 23.

What the mathematics does not license

Euclidean distance may ignore barriers. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Test scaled axes. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

Did You Know? In this chapter, the most important number is useful only together with its construction rule. Changing scaling, sample membership or validation design can change the numerical answer without changing the arithmetic.

5. Neighbourhood update

A small example makes the mechanism visible. The winner and nearby SOM nodes move toward the site.

A concrete situation

Early broad updates order the map.

What the mathematics does not license

Learning schedule shapes topology. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Record rate and radius. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

6. Topology preservation

This is where modelling choices begin to affect evidence. Nearby original sites should map to nearby nodes.

A concrete situation

A continuous layer stays continuous on the grid.

What the mathematics does not license

Folding can create false adjacency. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Measure topographic error. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

7. Site-to-node groups

The central mathematical idea is concrete. All sites with one winner form a metaspot.

A concrete situation

Fifty sparse beads share a node.

What the mathematics does not license

Group sizes vary strongly. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Map occupancy. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

8. Mean expression

Here the geometry becomes operational. Average gene expression summarizes a group.

A concrete situation

A node mean reduces noise.

What the mathematics does not license

Sparse signals can vanish. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Compare detection rates. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

9. Maximum expression

This step deserves a careful pause. The local maximum retains a strong rare measurement.

A concrete situation

One hotspot raises node value.

What the mathematics does not license

Outliers can dominate. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Use influence checks. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

10. Meta-expression mixture

The useful question is not only how to calculate, but what the calculation assumes. A weight combines maximum and mean.

A concrete situation

Gamma balances sensitivity and robustness.

What the mathematics does not license

One default cannot suit every gene. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Sweep the weight. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

11. Adjustable resolution

A small example makes the mechanism visible. Node count controls compression and spatial detail.

A concrete situation

Doubling k preserves a smaller region.

What the mathematics does not license

More nodes cost more and may be empty. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Plot accuracy versus runtime. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

12. Node Gaussian process

This is where modelling choices begin to affect evidence. Meta-expression covariance depends on node distance.

A concrete situation

Close nodes share a squared-exponential kernel.

What the mathematics does not license

Condensed geometry approximates original distances. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Compare kernel matrices. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

13. Length-scale models

The central mathematical idea is concrete. Several candidate Gaussian ranges are fitted.

A concrete situation

The best likelihood ratio uses one of ten scales.

What the mathematics does not license

Selecting the maximum raises search opportunity. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Calibrate selection. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

14. Spatial and noise variance

Here the geometry becomes operational. The GP separates coordinate-linked and residual variability.

A concrete situation

A broad pattern has high spatial fraction.

What the mathematics does not license

Compression changes both components. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Compare several k values. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

15. Back-mapping patterns

This step deserves a careful pause. Node results are returned to original sites for visualization.

A concrete situation

All sites in one node share the summary.

What the mathematics does not license

Back-mapping cannot restore lost within-node detail. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Overlay raw values. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

16. Governance and learning

The useful question is not only how to calculate, but what the calculation assumes. Version SOM grid, seed, schedule, gamma, kernels and tests.

A concrete situation

A student condenses twelve sites to four nodes.

What the mathematics does not license

Speed is bought with approximation. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Write a SOMDE audit card. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

A seven-day practice lab

Day 1: define the decision, observation unit, X variables and Y or reference condition.
Day 2: reproduce the smallest worked example and check every dimension, unit and sign.
Day 3: build a transparent baseline and write what it assumes.
Day 4: fit the main method inside a training-only pipeline; record every parameter.
Day 5: challenge one assumption with a deliberately difficult dataset or operating regime.
Day 6: compare validation metrics with the practical workload created by errors.
Day 7: present one supported conclusion, one unresolved uncertainty and one safe next experiment.

Questions for a student, parent or mentor

Which part of the model is a mathematical definition, and which part is a choice?
What small calculation would catch a sign, scale or dimension error?
What evidence comes from training, validation and the untouched test?
What benign condition could look like a fault or prediction failure?
How would the answer change for another response, asset, school project or time period?
What should a human review before acting on the score?

A method-specific audit before deployment

SOMDE tackles scale by condensing many spatial locations before fitting gene-level models. A self-organizing map places nodes on a regular grid and adjusts their coordinate weights so nearby original sites map to nearby nodes. Each node becomes a metaspot representing a group of spatial measurements.

Node-level meta-expression combines the local maximum and mean, a deliberate response to sparse counts. The compression resolution controls the number of nodes and therefore the speed-detail trade-off. Fine structures smaller than a node may disappear, while too many nodes return the computational burden the map was meant to reduce.

On the condensed map, a Gaussian process decomposes meta-expression into spatial and nonspatial variance. Several fixed length scales are fitted, the best likelihood ratio is retained and a fraction of spatial variation summarizes effect size. Model selection across scales must be incorporated into validation rather than treated as a free search.

A small exercise assigns twelve sites to four SOM nodes, calculates maximum, mean and a weighted meta-expression for one gene, then constructs a four-by-four Gaussian kernel. Change the grid resolution or maximum-mean weight to see exactly which local hotspot survives compression.

Frequently asked questions

What is SOMDE?

SOMDE is a scalable spatially variable gene method that first condenses locations with a self-organizing map and then fits Gaussian-process models to node-level meta-expression.

What does the self-organizing map preserve?

Its competitive and neighbourhood updates aim to preserve the topology of spatial coordinates on an adjustable regular grid.

What is a metaspot?

It is a SOM node representing original sites assigned to that node, with a node coordinate and a gene meta-expression summary.

Why combine maximum and mean expression?

The published method uses both to reduce the chance that sparse local signal is washed out by averaging, with a tunable combination weight.

Where does speed come from?

Gaussian-process covariance is fitted on the smaller node map rather than all original locations, greatly reducing matrix size.

What is the main limitation?

Compression can merge fine boundaries, rare hotspots or heterogeneous cells, so resolution and summary sensitivity must be reported.

Primary source and fact-check note

Minsheng Hao, Kui Hua and Xuegong Zhang, “SOMDE: a scalable method for identifying spatially variable genes with self-organizing map,” Bioinformatics 37, 4392–4398 (2021), DOI 10.1093/bioinformatics/btab471. Definitions and core claims in this article were checked against this primary publication. Calculations shown here were independently recomputed. Examples are educational illustrations; they are not reported industrial performance results.

Useful next reading

Keep the options open. Mathematics education supports clearer questions, transferable problem-solving skills and informed choices across science, engineering, computing, AI, statistics, design and many other careers. It does not guarantee intelligence, income, admission or employment; those outcomes depend on far more than one subject or model.

Extended laboratory: topic-specific validation and responsible use

Mask audit

Inspect fields of view and coordinate gaps.

SOM audit

Report grid, seeds and topographic error.

Occupancy audit

Map sites per node.

Summary audit

Sweep maximum-mean weight.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Resolution audit

Plot runtime and recovery versus k.

GP audit

Check length-scale search and null calibration.

Back-map audit

Overlay node summaries and raw sites.

Final claim

Compression trades detail for speed.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Mechanism workshop: calculate, challenge and transfer

Workshop 1: Spatial input sites

Begin with this mechanism: Each cell or spot has coordinates and gene expression. Work through the concrete case: Twenty thousand sites cover one section. Make the limitation visible: Coordinates may contain gaps or fields of view. Complete the evidence step: Plot the mask. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Workshop 2: SOM grid

Begin with this mechanism: A regular k-by-k node layout initializes the condensed map. Work through the concrete case: A ten-by-ten grid creates one hundred nodes. Make the limitation visible: Grid topology imposes structure. Complete the evidence step: Compare shapes and k. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Workshop 3: Node coordinate weights

Begin with this mechanism: Every node stores a learnable coordinate vector. Work through the concrete case: Weights begin across the tissue range. Make the limitation visible: Initialization affects convergence. Complete the evidence step: Fix seeds and restarts. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Workshop 4: Winner node

Begin with this mechanism: Each site maps to its closest coordinate weight. Work through the concrete case: One site selects node 23. Make the limitation visible: Euclidean distance may ignore barriers. Complete the evidence step: Test scaled axes. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Workshop 5: Neighbourhood update

Begin with this mechanism: The winner and nearby SOM nodes move toward the site. Work through the concrete case: Early broad updates order the map. Make the limitation visible: Learning schedule shapes topology. Complete the evidence step: Record rate and radius. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Workshop 6: Topology preservation

Begin with this mechanism: Nearby original sites should map to nearby nodes. Work through the concrete case: A continuous layer stays continuous on the grid. Make the limitation visible: Folding can create false adjacency. Complete the evidence step: Measure topographic error. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Workshop 7: Site-to-node groups

Begin with this mechanism: All sites with one winner form a metaspot. Work through the concrete case: Fifty sparse beads share a node. Make the limitation visible: Group sizes vary strongly. Complete the evidence step: Map occupancy. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Workshop 8: Mean expression

Begin with this mechanism: Average gene expression summarizes a group. Work through the concrete case: A node mean reduces noise. Make the limitation visible: Sparse signals can vanish. Complete the evidence step: Compare detection rates. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Workshop 9: Maximum expression

Begin with this mechanism: The local maximum retains a strong rare measurement. Work through the concrete case: One hotspot raises node value. Make the limitation visible: Outliers can dominate. Complete the evidence step: Use influence checks. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Workshop 10: Meta-expression mixture

Begin with this mechanism: A weight combines maximum and mean. Work through the concrete case: Gamma balances sensitivity and robustness. Make the limitation visible: One default cannot suit every gene. Complete the evidence step: Sweep the weight. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Workshop 11: Adjustable resolution

Begin with this mechanism: Node count controls compression and spatial detail. Work through the concrete case: Doubling k preserves a smaller region. Make the limitation visible: More nodes cost more and may be empty. Complete the evidence step: Plot accuracy versus runtime. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Workshop 12: Node Gaussian process

Begin with this mechanism: Meta-expression covariance depends on node distance. Work through the concrete case: Close nodes share a squared-exponential kernel. Make the limitation visible: Condensed geometry approximates original distances. Complete the evidence step: Compare kernel matrices. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Did You Know? A sophisticated adjustment can be algebraically correct and still answer the wrong scientific question when samples, labels, batches or validation boundaries are wrong.

Workshop 13: Length-scale models

Begin with this mechanism: Several candidate Gaussian ranges are fitted. Work through the concrete case: The best likelihood ratio uses one of ten scales. Make the limitation visible: Selecting the maximum raises search opportunity. Complete the evidence step: Calibrate selection. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Workshop 14: Spatial and noise variance

Begin with this mechanism: The GP separates coordinate-linked and residual variability. Work through the concrete case: A broad pattern has high spatial fraction. Make the limitation visible: Compression changes both components. Complete the evidence step: Compare several k values. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Workshop 15: Back-mapping patterns

Begin with this mechanism: Node results are returned to original sites for visualization. Work through the concrete case: All sites in one node share the summary. Make the limitation visible: Back-mapping cannot restore lost within-node detail. Complete the evidence step: Overlay raw values. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.

Workshop 16: Governance and learning

Begin with this mechanism: Version SOM grid, seed, schedule, gamma, kernels and tests. Work through the concrete case: A student condenses twelve sites to four nodes. Make the limitation visible: Speed is bought with approximation. Complete the evidence step: Write a SOMDE audit card. Write the coordinates, pair distances, graph weights, node groups, kernels, covariance matrices, eigenvalues, Fourier coefficients, permutation statistics, module correlations or reconstruction errors with named units. Calculate one small example by hand, then vary one assumption and explain which result moves and why. Reserve complete tissues, donors, genes, fields of view or regions outside tuning when testing transfer. Keep measurement, fitted representation, statistical evidence, biological interpretation and decision on separate notebook lines so a student, parent or reviewer can audit the mathematics.