FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteEvidence version preferred25 / 162 · 2e4275d0310efc09

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

1. Introduction

…tasets to show the practical applications of standard LP models. Moreover, we compare the similarity in prediction trends among the models and the specific network attributes that contribute to effective link prediction, before underscoring the role of LP in addressing the formidable challenges prevalent in biological systems, ranging from noise, bias, and data sparseness to interpretability. We conclude the review with an exploration of the essential characteristics expected from future LP models, poised to advance our comprehension of the intricate interactions governing biological systems.

↓ 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

Biological systems are characterized by densely interconnected entities, with genes, proteins, and metabolites participating in multiple overlapping interactions that give rise to highly connected hub nodes. Such structural properties distinguish biological networks from the sparser, more uniformly connected graphs typically assumed in general-purpose link prediction benchmarks.

Passages supplied to the Evidence version

Neural Multi-Hop Reasoning With Logical Rules on Biomedical Knowledge Graphs

From a machine learning perspective, reasoning on biomedical KGs presents new challenges for existing approaches due to the unique structural characteristics of the KGs. One challenge arises from the highly coupled nature of entities in biological systems that leads to many high-degree entities that…

Read full passage excerpt

From a machine learning perspective, reasoning on biomedical KGs presents new challenges for existing approaches due to the unique structural characteristics of the KGs. One challenge arises from the highly coupled nature of entities in biological systems that leads to many high-degree entities that are themselves densely linked. For example, as illustrated in Figure 1(a) , genes interact abundantly among themselves.

No Evidence

Same model and position · no retrieved passages · first output

Not preferred in this comparison

Link prediction (LP) has become a central computational task in network biology, where the goal is to infer missing or future interactions from the observed topology of a biological network. Because experimentally mapping interactomes, transcriptional regulatory circuits, and disease-gene associations remains costly and incomplete, computational approaches that prioritize candidate links have become essential to biological discovery.

Blind model judgment

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

Read the original model judgment

Output A contains unsupported claims: it introduces 'Link prediction (LP) has become a central computational task in network biology' and discusses 'experimentally mapping interactomes, transcriptional regulatory circuits, and disease-gene associations remains costly and incomplete' and 'computational approaches that prioritize candidate links have become essential to biological discovery' — none of which are supported by the supplied source. The source only discusses reasoning on biomedical KGs and challenges from highly coupled entities/high-degree nodes, not LP as a central task, experimental cost, or essential discovery tools. Output B stays grounded in the source's specific content: densely interconnected entities, genes, proteins, metabolites, hub nodes, and the contrast with general-purpose benchmarks — all directly supported by the source's discussion of 'highly coupled nature of entities,' 'high-degree entities,' 'densely linked,' and 'genes interact abundantly among themselves.' The structural comparison to 'sparser, more uniformly connected graphs' is a reasonable inference from the source's emphasis on biological network distinctiveness. Both outputs are coherent academic paragraphs, but Output A adds substantial unsupported factual claims while Output B maintains closed-book support.

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.