FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteEvidence version preferred27 / 162 · e21a23f73a3fce55

A Review of Link Prediction Applications in Network Biology

Quantitative biology · 2312.01275v1

FLOWING EVIDENCE BENCHMARK

How do we tell whether Evidence helps?

At the same writing position, with the same model and task, how does supplying retrieved paper passages change the first output? We compare matched versions and retain ties, unusable outputs and incomplete reviews.

SAME DRAFT · EVIDENCE ON OR OFF

01 · FIXED WRITING POSITION

MANUSCRIPT

Same writing position ▌

Matched autocomplete at the same draft position

02 · TWO MATCHED INPUTS

SHARED BY BOTH

Manuscript context, model, task and prompt

A · With Evidence

Retrieved paper passages supplied

B · Without Evidence

No retrieved passages supplied

LLM

Same model and version

A → first output

B → first output

Blind judge agent

First outputs are anonymized as X and Y

Continuation review: accuracy, fit to the writing task and usability
Returns: X preferred / tie / Y preferred / both unusable

Order check: X / Y → Y / X

HOW DOES BLIND JUDGING WORK?

① Anonymize both outputs
The judge sees the same draft and both first outputs without knowing which received Evidence.

② Compare and swap order
The judge applies task-specific criteria in X/Y and then Y/X order.

③ Review disagreements
A third pass resolves disagreements. Incomplete reviews remain in the denominator.

How are autocomplete positions stratified?

Before seeing generation outcomes, we check whether a retrieved passage contains a specific proposition that directly supports the next writing move. Those positions appear in the left opportunity group; the rest are ordinary positions on the right. We select a balanced sample from admitted papers in each field. The 50/50 split is experimental, not a measure of how often either type occurs in writing.

AUTOCOMPLETE · 110

Biology, statistics and astrophysics

55 positions on each side; statistics uses the ten-paper rerun.

AUTOCOMPLETE · 52

Psychology and climate science

26 positions on each side; psychology includes nine papers and climate science four.

How are the table percentages calculated?

Across five fields, 51 of 81 left-column positions preferred the Evidence version. Ties, pairs where both versions were unusable and incomplete reviews remain in the denominator.

51Evidence version preferred
÷
81All positions in this group
=
63%Evidence preference in this group

Source contribution is a separate review: 16 of 22 Evidence wins entered into source review directly used retrieved papers; another 29 wins await review.

These are development-stage model judgments pending independent human review. They are not formal Benchmark conclusions and do not, on their own, establish causality.

Manuscript writing position

Text excerpt · not a PDF page

Research manuscript · excerpt

A Review of Link Prediction Applications in Network Biology

6. Challenges and Future Directions > 6.2. Interpretability

…wn interactions are used to train the models before reporting unknown associations ( muzio2021biological ) . Thus, the missing associations are likely to create bias in the networks and have an adverse effect on the subsequent predictions. The fact that the networks are not complete precludes us from deriving inferences from their known topological properties. For instance, the knowledge that signaling, transcriptional, and protein networks are scale-free ( barabasi2004network ; albert2005scale ) does not warrant that the sampled subnetworks are scale-free as well ( heath2009computational ) .

…ure symmetry and inversions. Along the same lines, the embedding strategy needs to be relevant in the context of a biological question. If nodes are embedded based on local network topology, then the prediction cannot be expected to capture global features. Biological information, such as the structure of proteins and protein complexes, the layout of signaling pathways, or the higher-order organization of organelles, etc. are likely to be lost ( ideker2017network ) . Moreover, the noise and sparseness in the biomedical datasets diminish the predictive capabilities of the embedding approaches.

↓ Next paragraph was generated here

The highlighted text is near the writing position. This is a layout of the evaluation input, not a PDF screenshot.

A matched comparison at the same writing positionJudgment: Evidence version preferred

Both arms use the same model, task and manuscript context. Only the retrieved passages differ. The first recorded output from each arm is shown below.

Evidence

Same model and position · retrieved passages supplied · first output

Preferred by blind review

Higher-order representations offer one route to this end: hypergraphlets applied to hypernetworks in which nodes are proteins and hyperedges encode protein–protein interactions, complexes, or signaling pathways can test whether such topology carries biological information beyond that recoverable from pairwise PPI networks.

Passages supplied to the Evidence version

Higher order molecular organisation as a source of biological function

We investigate biological hypernetworks in which nodes are proteins and hyperedges capture PPIs, protein complexes, or signaling pathways. The main aim is to check if the topology of these hypernetwork representations of the data carries biological information that goes beyond the information that c…

Read full passage excerpt

We investigate biological hypernetworks in which nodes are proteins and hyperedges capture PPIs, protein complexes, or signaling pathways. The main aim is to check if the topology of these hypernetwork representations of the data carries biological information that goes beyond the information that can be obtained from PPI networks. We use hypergraphlets in this investigation.

No Evidence

Same model and position · no retrieved passages · first output

Not preferred in this comparison

No continuation text was produced.

Model action：complete

Blind model judgment

The blinded model review preferred the Evidence version. Its original rationale appears below.

Read the original model judgment

Output A is empty when the task clearly requires content—a coherent academic paragraph to continue the manuscript. This makes it unusable. Output B provides a single coherent academic paragraph that directly draws on the supplied source about hypergraphlets and hypernetworks, connecting to the draft's themes of network embedding limitations and biological information loss. The paragraph is source-grounded: it uses the exact entities (hypergraphlets, hypernetworks, proteins, hyperedges, PPIs, protein complexes, signaling pathways) and purpose (testing whether topology carries biological information beyond pairwise PPI networks) from the source snippet. It naturally extends the draft's argument that local topology-based embeddings lose biological information, by proposing higher-order representations as a potential solution. No unsupported claims are introduced; the source's investigational framing ('can test whether') is preserved rather than overstated as a finding.

Source review confirmed that this continuation directly used specific, verifiable information from the retrieved paper.

The passages are input material; their presence does not mean an output used them or that they support every claim. Draft and source passages are excerpted; no simulated scores or PDF appear here.