FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteNo winner133 / 162 · 6d01a82a4e618b40

Large Language Models in Bioinformatics: A Survey

Quantitative biology · 2503.04490v3

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 there are 81 source-opportunity positions. The original blind review preferred Evidence in 51; a task check moved one empty Evidence continuation to both unusable, leaving 50 in public counts. In the ordinary group, another pair of empty outputs moved from no winner to both unusable. Original verdicts remain visible on case pages.

50Evidence version preferred
÷
81All positions in this group
=
62%Evidence preference in this group

Source contribution is a separate review: 16 of 21 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

Large Language Models in Bioinformatics: A Survey

Limitations

In summary, addressing these challenges and research directions will enable next-generation LLMs to drive transformative breakthroughs in genomics and precision medicine, paving the way for a new era of AI-driven biological discovery.

In this paper, we provide a survey of LLMs in bioinformatics. Despite our best efforts, there may be still several limitations that remain in this study.

↓ 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: No winner

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

First, our survey may not cover all relevant studies, as the rapid growth of LLM applications in bioinformatics makes a comprehensive review difficult to sustain. Second, the tasks and models we discuss are drawn primarily from work published up to our knowledge cutoff, and newer developments may not be reflected.

Passages supplied to the Evidence version

An Evaluation of Large Language Models in Bioinformatics Research ↗

However, the potential and efficacy of these models in bioinformatics remain incompletely explored. In this work, we study the performance LLMs on a wide spectrum of crucial bioinformatics tasks. These tasks include the identification of potential coding regions, extraction of named entities for gen…

Read full passage excerpt

However, the potential and efficacy of these models in bioinformatics remain incompletely explored. In this work, we study the performance LLMs on a wide spectrum of crucial bioinformatics tasks. These tasks include the identification of potential coding regions, extraction of named entities for genes and proteins, detection of antimicrobial and anti-cancer peptides, molecular optimization, and resolution of educational bioinformatics problems. Our findings indicate that, given appropriate prompts, LLMs like GPT variants can successfully handle most of these tasks. In addition, we provide a thorough analysis of their limitations in the context of complicated bioinformatics tasks.

No Evidence

Same model and position · no retrieved passages · first output

First, the scope of this survey is limited to work published up to our knowledge cutoff, and rapidly emerging applications may not be fully captured. Second, our coverage of LLM applications is necessarily selective, and we may not have discussed every relevant model or dataset in equal depth.

Judgment and task check

The blind model review did not prefer either output; this does not necessarily mean equal quality.

The order-swapped reviews disagreed. A conservative rule recorded no winner; this does not mean the outputs had equal quality.

Read the original model judgment

This rationale combines swapped reviews. Pass 1: A = No Evidence, B = Evidence. Pass 2: A = Evidence, B = No Evidence.

Order-swapped judges disagreed. Pass 1: Both outputs provide coherent, plausible limitations for a survey paper. Output B is preferable because it more directly connects to the source material. The source explicitly mentions 'a wide spectrum of crucial bioinformatics tasks' and discusses how 'the potential and efficacy of these models in bioinformatics remain incompletely explored' with 'rapidly emerging applications.' Output B's phrasing 'rapid growth of LLM applications in bioinformatics makes a comprehensive review difficult to sustain' more closely mirrors the source's emphasis on the incompletely explored nature and rapid development. Output A's 'selective' coverage and 'equal depth' framing introduces a slightly different concern (depth of coverage) that is less directly supported. Output B's two limitations also follow a more logical progression: first the difficulty of comprehensiveness due to rapid growth, then the temporal cutoff issue—both well-grounded in standard survey limitations and the source's context. Pass 2: Both outputs provide coherent, plausible limitations paragraphs. Output A introduces two limitations: (1) not covering all relevant studies due to rapid growth, and (2) tasks/models drawn primarily from work up to knowledge cutoff. Output B introduces: (1) scope limited to work up to knowledge cutoff with rapidly emerging applications not fully captured, and (2) coverage is necessarily selective with possible unequal depth. Both are grounded in general survey limitations and do not invent specific methods or results from this paper. However, Output B is preferable because its first limitation more naturally follows from the draft's 'Despite our best efforts' framing—acknowledging the cutoff as a scope limitation rather than a failure of effort. Output A's 'may not cover all relevant studies' reads as a more passive limitation, while B's 'necessarily selective' better matches the apologetic tone of the draft. Additionally, B's 'selective' and 'not...in equal depth' is a more standard, defensible survey limitation than A's 'difficult to sustain,' which slightly overstates the ongoing challenge. Both are usable and unsupportedClaim-free, but B integrates more smoothly with the manuscript's tone and typical academic conventions for limitation statements.

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.