WHY MATHEMATICS?

Why Mathematics? | CrossTalkeR Differential Networks: Outdegree, Indegree, Betweenness, PageRank and Log-Odds

How CrossTalkeR turns ligand-receptor tables into cell and cell-gene graphs, then uses centrality and probability-scale comparisons to identify structural changes without confusing network prominence with biological control.

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 cell-cell and cell-gene directed weighted graphs
Separate outgoing influence, incoming listening and path mediation
Interpret PageRank as a damped stationary distribution
Compare phenotypes with matched graphs, differences and log-odds

Chapter map

1. Input edge table
2. Cell-cell nodes
3. Cell-cell weights
4. Cell-gene nodes
5. Directed edge
6. Outdegree
7. Outgoing strength
8. Indegree
9. Incoming strength
10. Shortest paths
11. Betweenness
12. Damping
13. Stationary probability
14. Phenotype difference
15. PageRank log-odds
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. Input edge table

The central mathematical idea is concrete. Each row names ligand, receptor, sender, receiver and positive weight.

A concrete situation

One L-R edge goes A to B.

What the mathematics does not license

Garbage-in propagates into every graph metric. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Audit upstream inference. 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. Cell-cell nodes

Here the geometry becomes operational. Each annotated cell type becomes one graph node.

A concrete situation

A, B and C form three nodes.

What the mathematics does not license

Broad labels hide states. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Test annotation resolution. 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. Cell-cell weights

This step deserves a careful pause. Molecular edges can be counted or summed between cell nodes.

A concrete situation

A-to-B has ten pairs totaling 24.

What the mathematics does not license

Count and weight answer different questions. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Publish both. 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. Cell-gene nodes

The useful question is not only how to calculate, but what the calculation assumes. A gene in a cell type becomes a distinct node.

A concrete situation

A:L differs from B:L.

What the mathematics does not license

The graph grows quickly. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Keep identifiers explicit. 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. Directed edge

A small example makes the mechanism visible. Ligand nodes point toward receptor nodes.

A concrete situation

A:L points to B:R.

What the mathematics does not license

Direction is inferred from molecule roles. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Avoid temporal claims. 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. Outdegree

This is where modelling choices begin to affect evidence. Outgoing edge count describes the number of destinations.

A concrete situation

A has outdegree four.

What the mathematics does not license

Thresholding changes degree. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Run cutoff 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.

7. Outgoing strength

The central mathematical idea is concrete. Outgoing weights describe total inferred signal.

A concrete situation

A sends weight 18.

What the mathematics does not license

One giant edge can dominate. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Show distribution. 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. Indegree

Here the geometry becomes operational. Incoming edge count describes the number of sources.

A concrete situation

B has indegree five.

What the mathematics does not license

Dense graphs inflate counts. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Compare matched density. 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. Incoming strength

This step deserves a careful pause. Incoming weights summarize received signal.

A concrete situation

B receives weight 22.

What the mathematics does not license

Weights inherit upstream scale. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Normalize consistently. 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. Shortest paths

The useful question is not only how to calculate, but what the calculation assumes. Paths depend on converting weights into lengths.

A concrete situation

Strong edges become short routes.

What the mathematics does not license

The transform changes routing. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

State the edge-length 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.

11. Betweenness

A small example makes the mechanism visible. Path fractions through a node quantify mediation.

A concrete situation

C lies on half the shortest routes.

What the mathematics does not license

Ties and disconnected parts matter. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Report normalization. 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. Damping

This is where modelling choices begin to affect evidence. PageRank mixes following edges with random restart.

A concrete situation

Damping is 0.85.

What the mathematics does not license

The parameter changes prominence. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Sweep damping. 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. Stationary probability

The central mathematical idea is concrete. Iterated flow converges to node importance.

A concrete situation

B receives probability 0.3.

What the mathematics does not license

Probability is graph-relative. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Do not call it biological probability. 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. Phenotype difference

Here the geometry becomes operational. Matched node scores can be subtracted.

A concrete situation

Outdegree rises by two.

What the mathematics does not license

Missing nodes create artificial change. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Align node universes. 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. PageRank log-odds

This step deserves a careful pause. Probability-scale importance is compared through odds.

A concrete situation

Log-odds is positive for B.

What the mathematics does not license

Zeros require a rule. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Report pseudocounts. 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 inputs, weights, graph type, paths, damping and contrasts.

A concrete situation

A student solves a four-node PageRank.

What the mathematics does not license

Centrality is model-dependent. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Write a CrossTalkeR graph 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

CrossTalkeR begins after a ligand-receptor inference method has produced a table of sender, receiver, ligand, receptor and positive edge weight. It builds two directed weighted graphs. In the cell-cell graph, nodes are cell types and edges aggregate molecular interactions between them. In the cell-gene graph, nodes are cell-type-and-gene pairs, so a ligand in one cell type and the same gene in another remain distinct mathematical objects.

Four graph summaries answer four different questions. Outgoing degree or weight describes an influencer that sends many signals. Incoming degree or weight describes a listener. Betweenness centrality counts how often a node lies on shortest paths and highlights possible moderators. PageRank treats importance as the stationary probability of a random walk with damping, so a node can rank highly because important neighbours point to it even when its raw degree is modest.

Comparing phenotypes requires matched node and edge definitions. Influencer, listener and moderator scores can be differenced directly. PageRank values are probabilities that sum to one inside each graph, so CrossTalkeR compares them through log-odds rather than a raw subtraction. A large differential score can still reflect changes in upstream inference, cell-type composition, graph density or an absent node. Zero handling and pseudocounts belong in the report.

Network centrality does not establish biological control. Betweenness is sensitive to path definition, PageRank depends on damping and edge normalization, and aggregated edges hide molecular diversity. The best interpretation is structural prioritization: which cells or cell-gene pairs deserve closer inspection in a declared network. Keeping the original ligand-receptor table beside every graph score prevents a sophisticated visualization from outrunning its evidence.

Frequently asked questions

What is CrossTalkeR?

CrossTalkeR is a graph-analysis framework that accepts scored ligand-receptor interactions and constructs cell-cell or cell-gene networks for one phenotype or a comparison of two phenotypes.

What is the influencer score?

It summarizes outgoing connections or signal weight from a node, depending on the selected network representation.

What does betweenness centrality mean?

It measures how often a node lies on shortest paths between other nodes under the declared edge-length rule. It is a structural mediation score.

What is PageRank here?

PageRank is the stationary distribution of a damped random walk on the directed graph, giving higher importance to nodes reached from other important nodes.

Why use log-odds for differential PageRank?

PageRank values are probabilities constrained to sum to one. Log-odds compare relative prominence on a probability scale while making the transformation and zero handling explicit.

Does a high centrality score prove a cell controls the tissue?

No. It prioritizes nodes inside an inferred graph; upstream scoring, graph construction, cell abundance and validation determine the biological claim.

Primary source and fact-check note

James S. Nagai, Nils B. Leimkühler, Michael T. Schaub, Rebekka K. Schneider and Ivan G. Costa, “CrossTalkeR: analysis and visualization of ligand-receptor networks,” Bioinformatics 37(22), 4263–4265 (2021), DOI 10.1093/bioinformatics/btab370. 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

Input audit

Preserve the complete scored ligand-receptor edge table.

Graph audit

State node definitions, directions and edge aggregation.

Degree audit

Separate counts from weighted strengths.

Path audit

Declare weight-to-length conversion and disconnected handling.

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.

PageRank audit

Publish damping, restart and convergence.

Contrast audit

Align node universes before phenotype comparison.

Odds audit

Document zero handling and pseudocounts.

Final claim

Centrality prioritizes structure; it does not prove control.

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: Input edge table

Begin with this mechanism: Each row names ligand, receptor, sender, receiver and positive weight. Recreate the concrete case: One L-R edge goes A to B. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Garbage-in propagates into every graph metric. Complete the transfer step: Audit upstream inference. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 2: Cell-cell nodes

Begin with this mechanism: Each annotated cell type becomes one graph node. Recreate the concrete case: A, B and C form three nodes. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Broad labels hide states. Complete the transfer step: Test annotation resolution. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 3: Cell-cell weights

Begin with this mechanism: Molecular edges can be counted or summed between cell nodes. Recreate the concrete case: A-to-B has ten pairs totaling 24. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Count and weight answer different questions. Complete the transfer step: Publish both. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 4: Cell-gene nodes

Begin with this mechanism: A gene in a cell type becomes a distinct node. Recreate the concrete case: A:L differs from B:L. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: The graph grows quickly. Complete the transfer step: Keep identifiers explicit. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion 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: Directed edge

Begin with this mechanism: Ligand nodes point toward receptor nodes. Recreate the concrete case: A:L points to B:R. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Direction is inferred from molecule roles. Complete the transfer step: Avoid temporal claims. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 6: Outdegree

Begin with this mechanism: Outgoing edge count describes the number of destinations. Recreate the concrete case: A has outdegree four. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Thresholding changes degree. Complete the transfer step: Run cutoff sensitivity. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 7: Outgoing strength

Begin with this mechanism: Outgoing weights describe total inferred signal. Recreate the concrete case: A sends weight 18. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: One giant edge can dominate. Complete the transfer step: Show distribution. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 8: Indegree

Begin with this mechanism: Incoming edge count describes the number of sources. Recreate the concrete case: B has indegree five. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Dense graphs inflate counts. Complete the transfer step: Compare matched density. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion 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: Incoming strength

Begin with this mechanism: Incoming weights summarize received signal. Recreate the concrete case: B receives weight 22. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Weights inherit upstream scale. Complete the transfer step: Normalize consistently. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 10: Shortest paths

Begin with this mechanism: Paths depend on converting weights into lengths. Recreate the concrete case: Strong edges become short routes. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: The transform changes routing. Complete the transfer step: State the edge-length rule. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 11: Betweenness

Begin with this mechanism: Path fractions through a node quantify mediation. Recreate the concrete case: C lies on half the shortest routes. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Ties and disconnected parts matter. Complete the transfer step: Report normalization. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 12: Damping

Begin with this mechanism: PageRank mixes following edges with random restart. Recreate the concrete case: Damping is 0.85. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: The parameter changes prominence. Complete the transfer step: Sweep damping. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion 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: Stationary probability

Begin with this mechanism: Iterated flow converges to node importance. Recreate the concrete case: B receives probability 0.3. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Probability is graph-relative. Complete the transfer step: Do not call it biological probability. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 14: Phenotype difference

Begin with this mechanism: Matched node scores can be subtracted. Recreate the concrete case: Outdegree rises by two. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Missing nodes create artificial change. Complete the transfer step: Align node universes. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.

Workshop 15: PageRank log-odds

Begin with this mechanism: Probability-scale importance is compared through odds. Recreate the concrete case: Log-odds is positive for B. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Zeros require a rule. Complete the transfer step: Report pseudocounts. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion 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 inputs, weights, graph type, paths, damping and contrasts. Recreate the concrete case: A student solves a four-node PageRank. Write edge counts, weights, outdegrees, indegrees, strengths, path lengths, betweenness fractions, PageRank mass and log-odds with direction, denominators and graph or probability assumptions. Calculate the smallest non-trivial example by hand, then reproduce it in code. Challenge this limitation: Centrality is model-dependent. Complete the transfer step: Write a CrossTalkeR graph card. Change exactly one annotation, edge threshold, graph aggregation, path transform, damping value, gene universe, expression cutoff, cluster size, Normal scale, subunit rule, probability bin, summary statistic or combination weight. Explain which mathematical object changes, which conclusion survives and which claim must weaken. Keep observation, transformation, null evidence, graph summary, pathway interpretation and biological conclusion on separate notebook lines. Reserve whole donors, tissues or conditions outside tuning when testing transfer.