FlowingFlowing
← Back to all development cases
This case was judged by an anonymized AI reviewer and has not been independently reviewed by people; inspect the draft, cards, and both outputs to assess it.
AutocompleteWithout cards better1 / 280 · beu-8f58833fb268bb31

Deciphering genomic codes using advanced NLP techniques: a scoping review

Quantitative biology · 2411.16084v1

FLOWING EVIDENCE BENCHMARK

How do we compare writing with and without cards?

At the same writing position, the same model produces one first result with cards and one without. All 280 pairs enter the overall anonymized review, which assesses whether a decision not to continue fits the draft.

Overall: 138 with-card wins, 26 ties, 107 no-card wins, 5 pairs where neither was usable, and 4 inconclusive judgments. Both sides wrote text in 218 pairs; 48 had a deliberate stop and 14 included an invalid output.

SAME DRAFT · WITH OR WITHOUT CARDS

01 · FIXED WRITING POSITION

MANUSCRIPT

Same writing position ▌

Matched autocomplete at the same draft position

02 · TWO MATCHED VERSIONS

SHARED BY BOTH

Draft, writing task, and generation model

A · With cards

Relevant literature cards supplied

B · Without cards

No literature cards supplied

Generation model

Same model and version

A → first output

B → first output

Anonymized AI judge

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

Review again with the answer order swapped

HOW DOES BLIND JUDGING WORK?

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

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 positions and cards selected?

We selected 40 writing positions in each of seven fields, for 280 comparisons. The position and relevant literature cards were set before each comparison, so both versions began from the same draft.

The main result retains every frozen position, including places where cards might not add a useful next move. This avoids selecting cases based on their outputs, but 49% describes the whole cohort, not a win rate where evidence clearly had an opportunity to help. This cohort did not record that opportunity label for each position before generation, so those two groups cannot yet be separated reliably.

CARD USE · 140/280

Reported card use in continuations with text

The model reports whether a card affected its continuation. In 140 of 280 tests it produced continuation text and reported card use. The previous count of 150 included 10 cases without continuation text and has been corrected.

COMPARISON · 70 / 7 / 30

Which written version was better?

Of the 107 reported-use cases where both versions produced text, the blinded AI judge preferred the with-card version 70 times, found 7 about the same, and preferred the without-card version 30 times.

How are the two table percentages calculated?

ADOPTION COVERAGE

140reported use with text
÷
280all positions
=
50%

WITH-CARD VERSION PREFERRED

70with cards better
÷
107comparable tests
=
65%

The percentages use different groups: 50% counts continuations with text reporting card use among all 280 tests; 65% covers the 107 such cases with text on both sides. The generating model reports card use, while an anonymized AI judge compares writing quality.

Card use is self-reported by the generator. The anonymized AI judge sees the draft and source snippets, so source-aware preference remains possible; there was no independent human review. The overall result describes these 280 frozen positions, not a win probability for any writing task.

Manuscript writing position

Text excerpt · not a PDF page

Research manuscript · excerpt

Deciphering genomic codes using advanced NLP techniques: a scoping review

1 Introduction

…n base pairs, contains information critical for understanding biological processes and disease mechanisms [ 3 ] . Traditional methods like Sanger sequencing, next-generation sequencing (NGS), and alignment-based approaches focus on generating and aligning sequence data but often fall short in interpreting large, complex genomic datasets, particularly for identifying regulatory regions and intricate patterns [ 4 ] . NLP and LLMs provide a scalable approach beyond raw sequencing, enabling efficient analysis, the discovery of regulatory regions, and deeper insights into genetic variation [ 5 ] .

…latory annotations. Tokenization involves converting raw genomic sequences into a format suitable for analysis, making the data more accessible for computational models [ 5 ] . With their advanced attention mechanisms, transformer architectures capture complex contextual relationships within the data, providing deeper insights into genomic structures [ 6 ] . Finally, predictive modeling uses the preprocessed data to identify critical regulatory elements such as transcription-factor binding sites, enhancer-promoter interactions, chromatin accessibility, and gene expression patterns [ 7 , 8 ] .

↓ 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 positionOverall AI review: Without cards better

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.

Read overall AI review rationale

A = with cards · B = without cards

Both outputs provide valid continuations that transition from the described advances to challenges/limitations. However, Output B is more usable for this draft position. The draft has established a clear structure: first paragraph introduces NLP/LLMs as a scalable approach for genomic analysis; second paragraph details three methodological components (tokenization, transformers, predictive modeling) with citations. The logical next move is to acknowledge limitations before potentially moving to future directions or conclusions. Output B identifies three concrete, well-established challenges in the field: (1) data scarcity, (2) interpretability, and (3) computational requirements. These map directly onto known issues in applying deep learning to genomics and maintain thematic coherence with the preceding text. Output A mentions only two challenges and introduces 'difficulty of transferring models trained on one genomic context to another'—a more specific claim that, while plausible, is less standardly cited and slightly narrows the scope. Output B's 'interpretability of deep learning models' directly connects to the draft's emphasis on 'interpreting large, complex genomic datasets' and 'deeper insights' mentioned earlier, creating better argumentative closure. The 'computational resources' point also naturally follows from discussion of 'large-scale models.' Both outputs are insertable without repair, but B's triad of challenges is more comprehensive and better aligned with the draft's established scope and terminology.

Evidence

Same model and position · retrieved passages supplied · first output

Not preferred in this comparison

Despite these advances, the application of NLP and LLMs to genomic data remains constrained by challenges that limit biological interpretation, including the scarcity of large, well-annotated datasets and the difficulty of transferring models trained on one genomic context to another.

Passages supplied to the Evidence version

Card 1 · Modelling the conditional regulatory activity of methylated and bivalent promoters

BACKGROUND: Predictive modelling of gene expression is a powerful framework for the in silico exploration of transcriptional regulatory interactions through the integration of high-throughput -omics data. A major limitation of previous approaches is their inability to handle conditional interactions…

Read full passage excerpt

BACKGROUND: Predictive modelling of gene expression is a powerful framework for the in silico exploration of transcriptional regulatory interactions through the integration of high-throughput -omics data. A major limitation of previous approaches is their inability to handle conditional interactions that emerge when genes are subject to different regulatory mechanisms. Although chromatin immunoprecipitation-based histone modification data are often used as proxies for chromatin accessibility, the association between these variables and expression often depends upon the presence of other epigenetic markers (e.g. DNA methylation or histone variants). These conditional interactions are poorly handled by previous predictive models and reduce the reliability of downstream biological inference. RESULTS: We have previously demonstrated that integrating both transcription factor and histone modification data within a single predictive model is rendered ineffective by their statistical redundan …

Card 2 · Chromatin Loops as Allosteric Modulators of Enhancer-Promoter Interactions

The classic model of eukaryotic gene expression requires direct spatial contact between a distal enhancer and a proximal promoter. Recent Chromosome Conformation Capture (3C) studies show that enhancers and promoters are embedded in a complex network of looping interactions. Here we use a polymer mo…

Read full passage excerpt

The classic model of eukaryotic gene expression requires direct spatial contact between a distal enhancer and a proximal promoter. Recent Chromosome Conformation Capture (3C) studies show that enhancers and promoters are embedded in a complex network of looping interactions. Here we use a polymer model of chromatin fiber to investigate whether, and to what extent, looping interactions between elements in the vicinity of an enhancer-promoter pair can influence their contact frequency. Our equilibrium polymer simulations show that a chromatin loop, formed by elements flanking either an enhancer or a promoter, suppresses enhancer-promoter interactions, working as an insulator. A loop formed by elements located in the region between an enhancer and a promoter, on the contrary, facilitates their interactions. We find that different mechanisms underlie insulation and facilitation; insulation occurs due to steric exclusion by the loop, and is a global effect, while facilitation occurs due to …

Card 3 · Integrative Modeling of eQTLs and Cis-Regulatory Elements Suggests Mechanisms Underlying Cell Type Specificity of eQTLs

Genetic variants in cis-regulatory elements or trans-acting regulators frequently influence the quantity and spatiotemporal distribution of gene transcription. Recent interest in expression quantitative trait locus (eQTL) mapping has paralleled the adoption of genome-wide association studies (GWAS)…

Read full passage excerpt

Genetic variants in cis-regulatory elements or trans-acting regulators frequently influence the quantity and spatiotemporal distribution of gene transcription. Recent interest in expression quantitative trait locus (eQTL) mapping has paralleled the adoption of genome-wide association studies (GWAS) for the analysis of complex traits and disease in humans. Under the hypothesis that many GWAS associations tag non-coding SNPs with small effects, and that these SNPs exert phenotypic control by modifying gene expression, it has become common to interpret GWAS associations using eQTL data. To fully exploit the mechanistic interpretability of eQTL-GWAS comparisons, an improved understanding of the genetic architecture and causal mechanisms of cell type specificity of eQTLs is required. We address this need by performing an eQTL analysis in three parts: first we identified eQTLs from eleven studies on seven cell types; then we integrated eQTL data with cis-regulatory element (CRE) data from th …

Card 4 · Granger causal inference on DAGs identifies genomic loci regulating transcription

of observations, Granger causality is a powerful approach for detecting predictive interactions between its variables. However, traditional Granger causal inference has limited utility in domains where the dynamics need to be represented as directed acyclic graphs (DAGs) rather than as a linear sequ…

Read full passage excerpt

of observations, Granger causality is a powerful approach for detecting predictive interactions between its variables. However, traditional Granger causal inference has limited utility in domains where the dynamics need to be represented as directed acyclic graphs (DAGs) rather than as a linear sequence, such as with cell differentiation trajectories. Here, we present GrID-Net, a framework based on graph neural networks with lagged message passing for Granger causal inference on DAG-structured systems. Our motivating application is the analysis of single-cell multimodal data to identify genomic loci that mediate the regulation of specific genes. To our knowledge, GrID-Net is the first single-cell analysis tool that accounts for the temporal lag between a genomic locus becoming accessible and its downstream effect on a target gene's expression. We applied GrID-Net on multimodal single-cell assays that profile chromatin accessibility (ATAC-seq) and gene expression (RNA-seq) in the same c …

Card 5 · How chromatin interactions shed light on interpreting non-coding genomic variants: opportunities and future direc-tions

Genomic variants, including copy number variants (CNVs) and genome-wide associa-tion study (GWAS) single nucleotide polymorphisms (SNPs), represent structural alterations that influence genomic diversity and disease susceptibility. While coding region variants have been extensively studied, non-codi…

Read full passage excerpt

Genomic variants, including copy number variants (CNVs) and genome-wide associa-tion study (GWAS) single nucleotide polymorphisms (SNPs), represent structural alterations that influence genomic diversity and disease susceptibility. While coding region variants have been extensively studied, non-coding and regulatory variants present significant challenges due to their potential impacts on gene regulation, which are often obscured by the complexity of the ge-nome. Chromatin interactions, which organize the genome spatially and regulate gene expression through enhancer-promoter contacts, predominantly occur in non-coding regions. Notably, more than 90% of enhancers, crucial for gene regulation, reside in these non-coding regions, underscor-ing their importance in interpreting the regulatory effects of CNVs and GWAS-associated SNPs. In this study, we integrate chromatin interaction data with CNV and GWAS data to uncover the functional implications of non-coding variants. By leveraging thi …

Model-reported card use

The model did not report that any card materially affected this completion.

No Evidence

Same model and position · no retrieved passages · first output

Preferred by blind review

Despite these advances, the application of NLP and LLMs to genomic data remains constrained by challenges that limit their broader adoption. These include the scarcity of large, well-annotated genomic datasets, the interpretability of deep learning models, and the computational resources required for training and deploying large-scale models.

Earlier card-use comparison

The earlier diagnostic compared only cases with reported card use and text on both sides; its scope differs from the overall AI review above.

This case does not have both with-card continuation text and a model-reported use record. It remains among all 280 tests but does not enter the reported-use comparison.

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.