WHY MATHEMATICS?

Why Mathematics? | CytoTalk Signalling Networks: Mutual Information, Network Propagation, Prize-Collecting Steiner Forests and Degree-Preserving Nulls

How CytoTalk searches an integrated two-cell-type graph for compact signaling subnetworks that connect active genes to ligand-receptor cross-talk while keeping optimization output separate from causal proof.

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

Construct cell-type-specific intracellular dependence networks
Join the networks through weighted ligand-receptor cross-talk
Turn expression and propagated proximity into prizes and costs
Optimize and validate a compact forest against degree-preserving nulls

Chapter map

1. Two cell-type expression matrices
2. Intracellular candidate edges
3. Mutual information
4. Network sparsification
5. Ligand-receptor bridge
6. Cross-talk weight
7. Seed set
8. Network propagation
9. Node prize
10. Edge cost
11. PCSF objective
12. Steiner nodes
13. Forest components
14. Solver and local alternatives
15. Degree-preserving null
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. Two cell-type expression matrices

The useful question is not only how to calculate, but what the calculation assumes. Separate gene expression is collected for a chosen sender and receiver pair.

A concrete situation

Macrophages and endothelial cells are analyzed.

What the mathematics does not license

Cluster labels may mix states. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Define both populations. 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. Intracellular candidate edges

A small example makes the mechanism visible. Possible gene-gene relationships are considered within each cell type.

A concrete situation

Genes A and B form a candidate edge.

What the mathematics does not license

Association is not regulation. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Keep direction claims modest. 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. Mutual information

This is where modelling choices begin to affect evidence. Dependence is measured beyond linear correlation.

A concrete situation

Two nonlinear gene profiles share information.

What the mathematics does not license

Estimation is sensitive to sample size and bins. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Use resampling. 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. Network sparsification

The central mathematical idea is concrete. Weak intracellular relationships are filtered.

A concrete situation

Top dependence edges remain.

What the mathematics does not license

Thresholds change connectivity. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Publish sensitivity. 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. Ligand-receptor bridge

Here the geometry becomes operational. Known molecular pairs join the two intracellular graphs.

A concrete situation

Ligand L connects receptor R.

What the mathematics does not license

Database context may be wrong. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Version pair evidence. 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. Cross-talk weight

This step deserves a careful pause. Cell-specific expression and self-talk penalties rank bridges.

A concrete situation

One L-R pair is favored across cell types.

What the mathematics does not license

A formula cannot observe secretion. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Preserve components. 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. Seed set

The useful question is not only how to calculate, but what the calculation assumes. Ligands and receptors act as anchors for downstream search.

A concrete situation

Five cross-talk pairs seed propagation.

What the mathematics does not license

Seed choice limits discoverable paths. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Report all eligible seeds. 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. Network propagation

A small example makes the mechanism visible. Seed influence diffuses across graph edges.

A concrete situation

Gene G lies close to receptor R.

What the mathematics does not license

Hubs can attract propagated mass. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Compare degree-aware controls. 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. Node prize

This is where modelling choices begin to affect evidence. Cell-specific activity and propagated closeness create reward.

A concrete situation

Active nearby gene G receives a large prize.

What the mathematics does not license

Scaling determines dominance. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Show prize distributions. 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. Edge cost

The central mathematical idea is concrete. Strong functional relationships become cheaper connections.

A concrete situation

A reliable edge has low cost.

What the mathematics does not license

Cost transforms can exaggerate small differences. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

State the transform. 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. PCSF objective

Here the geometry becomes operational. Collected prizes compete with paid edge costs and component penalties.

A concrete situation

A compact forest beats a sprawling graph.

What the mathematics does not license

Parameter values define compactness. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Write the objective. 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. Steiner nodes

This step deserves a careful pause. Low-prize connectors may enter because they link valuable terminals.

A concrete situation

One bridge gene connects two modules.

What the mathematics does not license

A connector is not necessarily active. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Label Steiner roles. 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. Forest components

The useful question is not only how to calculate, but what the calculation assumes. The solution may contain several disconnected signaling modules.

A concrete situation

Two pathways form separate trees.

What the mathematics does not license

Component penalties control splitting. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Sweep the forest parameter. 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. Solver and local alternatives

A small example makes the mechanism visible. An optimization procedure returns a high-scoring forest.

A concrete situation

Several near-optimal networks exist.

What the mathematics does not license

One optimum can hide uncertainty. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Retain alternative solutions. 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. Degree-preserving null

This is where modelling choices begin to affect evidence. Randomized graphs keep node degrees but shuffle specific connections.

A concrete situation

One thousand null forests form a score distribution.

What the mathematics does not license

The null cannot repair biased expression. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Report empirical tails. 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 central mathematical idea is concrete. Version MI estimation, thresholds, seeds, propagation, prizes, costs and nulls.

A concrete situation

A student solves a six-node prize-cost graph.

What the mathematics does not license

The forest is a prioritized hypothesis. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Write a CytoTalk 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

CytoTalk constructs an integrated network for a chosen pair of cell types. Within each cell type, gene-gene edges are weighted using mutual information, a measure of statistical dependence that can capture nonlinear association. The two intracellular networks are joined through known ligand-receptor edges. This creates a large search graph containing plausible intracellular routes and intercellular crossings.

Ligand-receptor cross-talk weights favor cell-type-specific expression and penalize patterns more consistent with self-talk. Network propagation measures how close other genes are to ligand or receptor seeds. Node prizes then combine cell-specific activity with this propagated closeness, while edge costs reflect the strength or probability of functional interaction. Each transformation changes which routes are inexpensive enough to enter the final network.

The prize-collecting Steiner forest problem selects a subnetwork that gains prizes by including informative nodes while paying costs for connecting them. A forest may contain several components, and tuning parameters control the trade-off between coverage and compactness. The optimizer solves the declared combinatorial objective; it does not guarantee the recovered edges are the unique biochemical pathway.

CytoTalk evaluates network significance against degree-preserving randomized networks. Preserving degree is important because hubs are easier to connect by chance. Yet a randomized null cannot remove batch effects, annotation errors or missing biology from the original graph. The responsible conclusion is a prioritized signaling-network hypothesis whose nodes and paths should be checked with spatial data, receptor perturbation, protein measurements and independent samples.

Frequently asked questions

What is CytoTalk?

CytoTalk constructs a cell-type-specific signaling network between two cell types from an integrated intracellular and ligand-receptor graph.

How are intracellular edges weighted?

The method uses mutual information between genes to represent statistical dependence within each cell type.

What does network propagation do?

It spreads influence from ligand and receptor seeds through the integrated graph to quantify how close other genes are to potential cross-talk.

What is a prize-collecting Steiner forest?

It selects one or more subnetworks that collect valuable nodes while paying edge costs, balancing biological coverage against compactness.

Why use degree-preserving randomized networks?

They keep the hub structure while disrupting specific connections, creating a stronger null for testing whether the selected network score is unusual.

Does the selected forest prove a biochemical pathway?

No. It is an optimized hypothesis based on expression, prior pairs and graph weights; perturbation, spatial and protein evidence remain necessary.

Primary source and fact-check note

Yuxuan Hu, Tao Peng, Lin Gao and Kai Tan, “CytoTalk: De novo construction of signal transduction networks using single-cell transcriptomic data,” Science Advances 7(16), eabf1356 (2021), DOI 10.1126/sciadv.abf1356. 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

Population audit

Define the two cell types and sample boundaries.

MI audit

Validate dependence estimates with resampling.

Bridge audit

Version ligand-receptor seeds and cross-talk weights.

Propagation audit

Check hub attraction and diffusion settings.

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.

Prize-cost audit

Show every scaling and transformation.

Optimization audit

Sweep PCSF parameters and keep alternatives.

Null audit

Use degree-preserving randomization and empirical tails.

Final claim

An optimized forest is a hypothesis, not a proven pathway.

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: Two cell-type expression matrices

Begin with this mechanism: Separate gene expression is collected for a chosen sender and receiver pair. Recreate the concrete case: Macrophages and endothelial cells are analyzed. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Cluster labels may mix states. Complete the transfer step: Define both populations. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.

Workshop 2: Intracellular candidate edges

Begin with this mechanism: Possible gene-gene relationships are considered within each cell type. Recreate the concrete case: Genes A and B form a candidate edge. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Association is not regulation. Complete the transfer step: Keep direction claims modest. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.

Workshop 3: Mutual information

Begin with this mechanism: Dependence is measured beyond linear correlation. Recreate the concrete case: Two nonlinear gene profiles share information. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Estimation is sensitive to sample size and bins. Complete the transfer step: Use resampling. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.

Workshop 4: Network sparsification

Begin with this mechanism: Weak intracellular relationships are filtered. Recreate the concrete case: Top dependence edges remain. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Thresholds change connectivity. Complete the transfer step: Publish sensitivity. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families 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: Ligand-receptor bridge

Begin with this mechanism: Known molecular pairs join the two intracellular graphs. Recreate the concrete case: Ligand L connects receptor R. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Database context may be wrong. Complete the transfer step: Version pair evidence. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.

Workshop 6: Cross-talk weight

Begin with this mechanism: Cell-specific expression and self-talk penalties rank bridges. Recreate the concrete case: One L-R pair is favored across cell types. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: A formula cannot observe secretion. Complete the transfer step: Preserve components. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.

Workshop 7: Seed set

Begin with this mechanism: Ligands and receptors act as anchors for downstream search. Recreate the concrete case: Five cross-talk pairs seed propagation. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Seed choice limits discoverable paths. Complete the transfer step: Report all eligible seeds. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.

Workshop 8: Network propagation

Begin with this mechanism: Seed influence diffuses across graph edges. Recreate the concrete case: Gene G lies close to receptor R. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Hubs can attract propagated mass. Complete the transfer step: Compare degree-aware controls. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families 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: Node prize

Begin with this mechanism: Cell-specific activity and propagated closeness create reward. Recreate the concrete case: Active nearby gene G receives a large prize. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Scaling determines dominance. Complete the transfer step: Show prize distributions. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.

Workshop 10: Edge cost

Begin with this mechanism: Strong functional relationships become cheaper connections. Recreate the concrete case: A reliable edge has low cost. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Cost transforms can exaggerate small differences. Complete the transfer step: State the transform. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.

Workshop 11: PCSF objective

Begin with this mechanism: Collected prizes compete with paid edge costs and component penalties. Recreate the concrete case: A compact forest beats a sprawling graph. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Parameter values define compactness. Complete the transfer step: Write the objective. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.

Workshop 12: Steiner nodes

Begin with this mechanism: Low-prize connectors may enter because they link valuable terminals. Recreate the concrete case: One bridge gene connects two modules. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: A connector is not necessarily active. Complete the transfer step: Label Steiner roles. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families 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: Forest components

Begin with this mechanism: The solution may contain several disconnected signaling modules. Recreate the concrete case: Two pathways form separate trees. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Component penalties control splitting. Complete the transfer step: Sweep the forest parameter. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.

Workshop 14: Solver and local alternatives

Begin with this mechanism: An optimization procedure returns a high-scoring forest. Recreate the concrete case: Several near-optimal networks exist. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: One optimum can hide uncertainty. Complete the transfer step: Retain alternative solutions. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.

Workshop 15: Degree-preserving null

Begin with this mechanism: Randomized graphs keep node degrees but shuffle specific connections. Recreate the concrete case: One thousand null forests form a score distribution. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: The null cannot repair biased expression. Complete the transfer step: Report empirical tails. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.

Workshop 16: Governance and learning

Begin with this mechanism: Version MI estimation, thresholds, seeds, propagation, prizes, costs and nulls. Recreate the concrete case: A student solves a six-node prize-cost graph. Write mutual-information weights, ligand-receptor bridges, propagated closeness, node prizes, edge costs, forest components, objective values and null scores with dimensions, direction and normalization. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: The forest is a prioritized hypothesis. Complete the transfer step: Write a CytoTalk audit card. Change exactly one annotation, threshold, subunit, permutation block, resource, score direction, candidate universe, specificity denominator, propagation setting, prize-cost scale or forest penalty. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep measurement, model output, null evidence, interpretation and decision on separate notebook lines. Reserve complete donors, tissues, conditions or interaction families outside tuning when testing transfer.