WHY MATHEMATICS?

Why Mathematics? | NICHES Single-Cell Signal Space: Expression Products, Niche Vectors, Sampling and Low-Dimensional Embeddings

How NICHES represents each cell pair or cellular microenvironment as a quantitative signal-space vector so hidden heterogeneity can be studied without treating a cluster average as every cell.

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

Build mechanism-by-pair matrices from sender and receiver expression
Distinguish a cell-pair signal from a receiver niche vector
Control pair sampling and rare-pattern coverage
Interpret embeddings as approximate maps of high-dimensional signal space

Chapter map

1. Mechanism rows
2. Sender cell
3. Receiver cell
4. Ordered-pair column
5. Product score
6. Complex subunits
7. Cell-pair matrix
8. Sender neighbourhood
9. Mean ligand input
10. Niche product
11. Pair sampling
12. Inclusion stability
13. Feature scaling
14. Dimension reduction
15. Marker mechanism
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. Mechanism rows

Here the geometry becomes operational. Each row names a ligand-receptor mechanism and subunits.

A concrete situation

Row one is L1-R1.

What the mathematics does not license

Resource coverage limits discovery. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Version the list. 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. Sender cell

This step deserves a careful pause. One observed cell supplies ligand values.

A concrete situation

Cell A expresses L1 at two.

What the mathematics does not license

Dropout can hide ligands. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Retain 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.

3. Receiver cell

The useful question is not only how to calculate, but what the calculation assumes. One observed cell supplies receptor values.

A concrete situation

Cell B expresses R1 at three.

What the mathematics does not license

Noise is cell-specific. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Use quality filters. 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. Ordered-pair column

A small example makes the mechanism visible. A-to-B is one high-dimensional observation.

A concrete situation

The column contains 2,000 mechanisms.

What the mathematics does not license

Pair counts grow quadratically. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Estimate scale. 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. Product score

This is where modelling choices begin to affect evidence. Each entry multiplies ligand and receptor expression.

A concrete situation

Two times three gives six.

What the mathematics does not license

Products depend on normalization. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Freeze transformation. 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. Complex subunits

The central mathematical idea is concrete. Custom mechanisms may contain multiple components.

A concrete situation

L has two required chains.

What the mathematics does not license

Aggregation rules change bottlenecks. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

List and test the rule. 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. Cell-pair matrix

Here the geometry becomes operational. Mechanism rows cross sampled pair columns.

A concrete situation

2,000 by 50,000 values.

What the mathematics does not license

Dense storage can be costly. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Use sparse formats carefully. 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. Sender neighbourhood

This step deserves a careful pause. A receiver is assigned surrounding or associated senders.

A concrete situation

Five spatial neighbours enter.

What the mathematics does not license

Radius and graph rules define exposure. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Sweep neighbourhoods. 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. Mean ligand input

The useful question is not only how to calculate, but what the calculation assumes. Sender ligand values are averaged for the niche.

A concrete situation

Mean L1 equals 1.4.

What the mathematics does not license

A mean hides one strong sender. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Show distribution and max. 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. Niche product

A small example makes the mechanism visible. Mean ligand input multiplies receiver receptor value.

A concrete situation

1.4 times three equals 4.2.

What the mathematics does not license

It is an exposure proxy. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Avoid probability language. 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. Pair sampling

This is where modelling choices begin to affect evidence. Unique sender-receiver pairs are drawn for each cell-type cross.

A concrete situation

Ten thousand of one million pairs are used.

What the mathematics does not license

Rare patterns may be missed. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Use stratified repeated samples. 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. Inclusion stability

The central mathematical idea is concrete. Repeated seeds test whether archetypes recur.

A concrete situation

A cluster appears in nine of ten runs.

What the mathematics does not license

Stability can reflect dominant structure. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Check rare-cell coverage. 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. Feature scaling

Here the geometry becomes operational. Mechanism rows may be normalized before embedding.

A concrete situation

Each row receives unit variance.

What the mathematics does not license

Scaling changes which mechanisms dominate. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Publish the choice. 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. Dimension reduction

This step deserves a careful pause. A high-dimensional pair vector maps to two or more coordinates.

A concrete situation

UMAP places similar pairs nearby.

What the mathematics does not license

Global distances may distort. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Compare PCA and neighbourhood metrics. 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. Marker mechanism

The useful question is not only how to calculate, but what the calculation assumes. Features enriched in one pair cluster are identified.

A concrete situation

L1-R1 marks a subgroup.

What the mathematics does not license

Post-selection tests can overstate evidence. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Validate on held-out data. 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

A small example makes the mechanism visible. Version resources, neighbourhoods, products, samples, scaling and embeddings.

A concrete situation

A student constructs a four-mechanism matrix.

What the mathematics does not license

A map is not the territory. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Write a NICHES signal-space 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

NICHES changes the unit of analysis from an average cell type to an individual ordered cell pair. Rows are ligand-receptor mechanisms. Columns are sender-receiver pairs. Each entry is the sender ligand expression multiplied by the receiver receptor expression. This simple product creates a signal-space vector for every pair. It preserves heterogeneity that disappears when all cells of one type are averaged before scoring.

A cellular niche is a related but different object. For one receiver, NICHES first averages ligand expression across a chosen set of neighbouring or associated sender cells, then multiplies that mean by receptor expression in the receiver. The resulting vector is a proxy for the extracellular signals the cell could sense. The sender set, spatial radius, graph-neighbour rule and weighting scheme define the niche; changing them changes the mathematical question.

The number of possible cell pairs can be enormous, so the single-cell workflow samples unique pairs for cell-type crosses when full enumeration is impractical. Sampling controls computation but creates Monte Carlo variability. Rare interaction archetypes may be missed if the sample is too small or poorly stratified. Repeated seeds, inclusion probabilities and stability checks help distinguish a robust pattern from a convenient draw.

Once the mechanism-by-pair or mechanism-by-niche matrix exists, standard single-cell tools can normalize, reduce dimensions, cluster and test markers. A low-dimensional embedding is a map that tries to preserve selected relationships among high-dimensional vectors. It is not the original geometry and does not prove that visible islands are distinct biological states. NICHES is strongest when the product scores, neighbourhood definition, sampling design and embedding parameters remain available beside the visualization.

Frequently asked questions

What is NICHES?

NICHES is a software framework that represents cell-cell signaling and cellular microenvironments in ligand-receptor signal space at single-cell resolution.

What is one column of the cell-pair matrix?

It is a vector across ligand-receptor mechanisms for one ordered sender-receiver pair.

How is a mechanism entry calculated?

The sender cell’s ligand expression is multiplied by the receiver cell’s receptor expression for that mechanism.

What is a niche vector?

For one receiver, mean ligand expression across a chosen sender neighbourhood is multiplied by the receiver’s receptor expression mechanism by mechanism.

Why sample cell pairs?

Full enumeration can be enormous. Sampling controls computation, but repeated seeds and coverage checks are needed to protect rare patterns.

What does an embedding show?

It is a low-dimensional approximation of similarity among high-dimensional signaling vectors, useful for exploration but not proof of discrete biological classes.

Primary source and fact-check note

Micha Sam Brickman Raredon, Junchen Yang, Neeharika Kothapalli, Wesley Lewis, Naftali Kaminski, Laura E. Niklason and Yuval Kluger, “Comprehensive visualization of cell-cell interactions in single-cell and spatial transcriptomics with NICHES,” Bioinformatics 39(1), btac775 (2023), DOI 10.1093/bioinformatics/btac775. 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

Resource audit

Version mechanism lists and all complex subunits.

Pair audit

Preserve sender, receiver and direction for every column.

Product audit

Record transformations before multiplication.

Niche audit

Declare neighbourhood membership and ligand averaging.

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.

Sampling audit

Publish inclusion design, seeds and rare-cell coverage.

Embedding audit

Store scaling, neighbours, distance and random seed.

Marker audit

Validate selected mechanisms on held-out data.

Final claim

Signal-space proximity is an exploratory representation, not causal proof.

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: Mechanism rows

Begin with this mechanism: Each row names a ligand-receptor mechanism and subunits. Recreate the concrete case: Row one is L1-R1. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Resource coverage limits discovery. Complete the transfer step: Version the list. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 2: Sender cell

Begin with this mechanism: One observed cell supplies ligand values. Recreate the concrete case: Cell A expresses L1 at two. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Dropout can hide ligands. Complete the transfer step: Retain detection rates. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 3: Receiver cell

Begin with this mechanism: One observed cell supplies receptor values. Recreate the concrete case: Cell B expresses R1 at three. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Noise is cell-specific. Complete the transfer step: Use quality filters. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 4: Ordered-pair column

Begin with this mechanism: A-to-B is one high-dimensional observation. Recreate the concrete case: The column contains 2,000 mechanisms. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Pair counts grow quadratically. Complete the transfer step: Estimate scale. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

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: Product score

Begin with this mechanism: Each entry multiplies ligand and receptor expression. Recreate the concrete case: Two times three gives six. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Products depend on normalization. Complete the transfer step: Freeze transformation. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 6: Complex subunits

Begin with this mechanism: Custom mechanisms may contain multiple components. Recreate the concrete case: L has two required chains. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Aggregation rules change bottlenecks. Complete the transfer step: List and test the rule. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 7: Cell-pair matrix

Begin with this mechanism: Mechanism rows cross sampled pair columns. Recreate the concrete case: 2,000 by 50,000 values. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Dense storage can be costly. Complete the transfer step: Use sparse formats carefully. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 8: Sender neighbourhood

Begin with this mechanism: A receiver is assigned surrounding or associated senders. Recreate the concrete case: Five spatial neighbours enter. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Radius and graph rules define exposure. Complete the transfer step: Sweep neighbourhoods. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

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: Mean ligand input

Begin with this mechanism: Sender ligand values are averaged for the niche. Recreate the concrete case: Mean L1 equals 1.4. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: A mean hides one strong sender. Complete the transfer step: Show distribution and max. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 10: Niche product

Begin with this mechanism: Mean ligand input multiplies receiver receptor value. Recreate the concrete case: 1.4 times three equals 4.2. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: It is an exposure proxy. Complete the transfer step: Avoid probability language. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 11: Pair sampling

Begin with this mechanism: Unique sender-receiver pairs are drawn for each cell-type cross. Recreate the concrete case: Ten thousand of one million pairs are used. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Rare patterns may be missed. Complete the transfer step: Use stratified repeated samples. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 12: Inclusion stability

Begin with this mechanism: Repeated seeds test whether archetypes recur. Recreate the concrete case: A cluster appears in nine of ten runs. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Stability can reflect dominant structure. Complete the transfer step: Check rare-cell coverage. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

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: Feature scaling

Begin with this mechanism: Mechanism rows may be normalized before embedding. Recreate the concrete case: Each row receives unit variance. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Scaling changes which mechanisms dominate. Complete the transfer step: Publish the choice. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 14: Dimension reduction

Begin with this mechanism: A high-dimensional pair vector maps to two or more coordinates. Recreate the concrete case: UMAP places similar pairs nearby. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Global distances may distort. Complete the transfer step: Compare PCA and neighbourhood metrics. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 15: Marker mechanism

Begin with this mechanism: Features enriched in one pair cluster are identified. Recreate the concrete case: L1-R1 marks a subgroup. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: Post-selection tests can overstate evidence. Complete the transfer step: Validate on held-out data. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 16: Governance and learning

Begin with this mechanism: Version resources, neighbourhoods, products, samples, scaling and embeddings. Recreate the concrete case: A student constructs a four-mechanism matrix. Write mechanism rows, ordered pairs, ligand values, receptor values, products, neighbourhood means, inclusion rates, scaled features and embedding distances with direction, units, denominators and assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in a spreadsheet or short program. Challenge this limitation: A map is not the territory. Complete the transfer step: Write a NICHES signal-space card. Change exactly one reference profile, normalization, regularization value, graph radius, spatial edge, shuffle unit, pair sample, anchor, bin size, feature scale, random seed, matrix rank, similarity threshold, root cell or pathway target set. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, transformation, optimization, null comparison, graph result, biological interpretation and validation on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.