Liang, Zhang, Yukun, Zhang, Yang, Tan, Fan, Jiang, Bozitao, Zhong, Xiaoran, Cheng, Liang, Hong, Mingchen, Li, Yuanxi, Yu · Oct 20, 2026
arXiv cs.AI: VenusGenome uses genomic context to represent microbial genes
Markdown · JSON · Original source
The authors of VenusGenome argue that a gene’s protein representation can be predicted in part from neighboring genes, so a model should learn from genomic context while distinguishing biological relationships from assembly conventions. Their gene-resolution masked autoencoder uses visible-only encoding and genome-aware attention that incorporates contig boundaries, relative strand orientations and gene positions. It is on over 1.7 million microbial genome assemblies, with frozen protein as gene-level inputs. Across genome-level phenotype prediction and gene-level tasks, its frozen representations have the highest mean performance among evaluated models on nine of eleven primary metrics. The authors also report that ablations support genome-aware attention and masked autoencoding, with further gains from higher masking ratios. An exploratory gene-input perturbation analysis links representation sensitivity to experimentally measured fitness defects. The results suggest a way to build microbial gene representations that use neighboring genes without treating separate contigs as adjacent; the abstract does not establish how performance transfers beyond the evaluated tasks.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Oct 20, 2026 (2026-10-20).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-10T00:24:45.376Z. First observed: 2026-11-05T05:00:25.247Z. Retrieved: 2026-11-10T00:23:58.836Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The performance comparison is limited to the evaluated models, tasks and eleven primary metrics; the abstract does not specify their full details. The authors describe the gene-input perturbation analysis as exploratory and say it links sensitivity to fitness defects; the abstract does not establish how strong or general that relationship is. The abstract reports results for frozen representations and does not establish transfer beyond the evaluated tasks.
Open question
What experiments would establish how well representation sensitivity tracks experimentally measured fitness defects beyond the exploratory perturbation analysis?
Source-stated
The authors say a gene's protein embedding is partly predictable from neighboring genes, motivating reconstruction from sparse genomic context.
a gene's protein embedding is partially predictable from the embeddings of neighboring genes in the genome
Source-stated
VenusGenome's genome-aware attention incorporates contig boundaries, relative strand orientations and gene positions.
genome-aware attention incorporating contig boundaries, relative strand orientations, and gene positions
Source-stated
The model is pretrained on over 1.7 million microbial genome assemblies using frozen protein embeddings as gene-level inputs.
We pretrain this model on a curated corpus of over 1.7 million microbial genome assemblies from isolate and metagenomic sources, using frozen protein embeddings as gene-level inputs.
Source-stated
The authors report the highest mean performance among evaluated models on nine of eleven primary metrics across genome-level phenotype prediction and gene-level tasks.
Across genome-level phenotype prediction and gene-level tasks, the resulting frozen representations achieve the highest mean performance among the evaluated models on nine of eleven primary metrics.
Source-stated
The authors report that an exploratory perturbation analysis links representation sensitivity to experimentally measured fitness defects.
An exploratory gene-input perturbation analysis further links representation sensitivity to experimentally measured fitness defects.
Source and contributor context
Liang, Zhang, Yukun, Zhang, Yang, Tan, Fan, Jiang, Bozitao, Zhong, Xiaoran, Cheng, Liang, Hong, Mingchen, Li, Yuanxi, Yu · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Pre-training
A model's initial training on a large dataset, before further adaptation for particular uses.
Embedding vector
A list of numbers that represents a piece of text, an image or another item so software can compare meanings.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 59928146-d7e6-4c96-8888-f0567739e927.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-10T00:24:45.376Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 0d70416534b1572ff81c38dc98590b78a41bd0c4d41c1042dca29fc71c7e69a2. Source text hash: cf110335b8492786bf2625ed9b71ac42e2e70deb15bdb52827d2d1d1ea775f8c.
Revision and source-check history
Version 1 · 2026-11-10T00:24:45.376Z
arXiv cs.AI: VenusGenome uses genomic context to represent microbial genes
The authors of VenusGenome argue that a gene’s protein representation can be predicted in part from neighboring genes, so a model should learn from genomic context while distinguishing biological relationships from assembly conventions. Their gene-resolution masked autoencoder uses visible-only encoding and genome-aware attention that incorporates contig boundaries, relative strand orientations and gene positions. It is on over 1.7 million microbial genome assemblies, with frozen protein as gene-level inputs. Across genome-level phenotype prediction and gene-level tasks, its frozen representations have the highest mean performance among evaluated models on nine of eleven primary metrics. The authors also report that ablations support genome-aware attention and masked autoencoding, with further gains from higher masking ratios. An exploratory gene-input perturbation analysis links representation sensitivity to experimentally measured fitness defects. The results suggest a way to build microbial gene representations that use neighboring genes without treating separate contigs as adjacent; the abstract does not establish how performance transfers beyond the evaluated tasks.
Original source
cs.AI · Article
Source details
Original source date: Oct 20, 2026 (2026-10-20).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-10T00:24:45.376Z. First observed: 2026-11-05T05:00:25.247Z. Retrieved: 2026-11-10T00:23:58.836Z.
Limits
The performance comparison is limited to the evaluated models, tasks and eleven primary metrics; the abstract does not specify their full details. The authors describe the gene-input perturbation analysis as exploratory and say it links sensitivity to fitness defects; the abstract does not establish how strong or general that relationship is. The abstract reports results for frozen representations and does not establish transfer beyond the evaluated tasks.
Open question
What experiments would establish how well representation sensitivity tracks experimentally measured fitness defects beyond the exploratory perturbation analysis?
Source-stated
The authors say a gene's protein embedding is partly predictable from neighboring genes, motivating reconstruction from sparse genomic context.
a gene's protein embedding is partially predictable from the embeddings of neighboring genes in the genome
Source-stated
VenusGenome's genome-aware attention incorporates contig boundaries, relative strand orientations and gene positions.
genome-aware attention incorporating contig boundaries, relative strand orientations, and gene positions
Source-stated
The model is pretrained on over 1.7 million microbial genome assemblies using frozen protein embeddings as gene-level inputs.
We pretrain this model on a curated corpus of over 1.7 million microbial genome assemblies from isolate and metagenomic sources, using frozen protein embeddings as gene-level inputs.
Source-stated
The authors report the highest mean performance among evaluated models on nine of eleven primary metrics across genome-level phenotype prediction and gene-level tasks.
Across genome-level phenotype prediction and gene-level tasks, the resulting frozen representations achieve the highest mean performance among the evaluated models on nine of eleven primary metrics.
Source-stated
The authors report that an exploratory perturbation analysis links representation sensitivity to experimentally measured fitness defects.
An exploratory gene-input perturbation analysis further links representation sensitivity to experimentally measured fitness defects.
Source and contributor context
Liang, Zhang, Yukun, Zhang, Yang, Tan, Fan, Jiang, Bozitao, Zhong, Xiaoran, Cheng, Liang, Hong, Mingchen, Li, Yuanxi, Yu · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Pre-training
A model's initial training on a large dataset, before further adaptation for particular uses.
Embedding vector
A list of numbers that represents a piece of text, an image or another item so software can compare meanings.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 59928146-d7e6-4c96-8888-f0567739e927.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-10T00:24:45.376Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 0d70416534b1572ff81c38dc98590b78a41bd0c4d41c1042dca29fc71c7e69a2. Source text hash: cf110335b8492786bf2625ed9b71ac42e2e70deb15bdb52827d2d1d1ea775f8c.
uncertain · 2026-11-10T00:24:45.376Z · anthropic/claude-opus-5.5
Publication: 59928146-d7e6-4c96-8888-f0567739e927
attribution — pass: The reading attributes the argument and results to the VenusGenome authors in an arXiv paper, and the title names the arXiv cs.AI listing. Reported results are framed as the authors' findings, not established facts.
We therefore introduce VenusGenome, a gene-resolution masked autoencoder
summary — uncertain: The summary is faithful but hard for a general reader to follow. Central terms are left unexplained: masked autoencoder, contig, strand orientation, protein embeddings, frozen representations, ablations, phenotype and fitness defects. The key payoff, avoiding treating separate contigs as adjacent, depends on knowing what a contig is. The stakes, better prediction of microbial function, are also not stated plainly. Smallest repair: briefly explain these terms or ensure tooltips exist, for example that contigs are separately assembled DNA fragments. Also restore the comparison against BERT-style masked modeling in the ablation sentence.
the ordering of separate contigs does not imply adjacency and strand labels depend on the reference orientation
date — pass: The originalPublished date of 2026-10-20 matches the v1 submission date. The 2026-11-05 feed date is correctly not used.
[Submitted on 20 Oct 2026]
support — pass: Each claim matches the abstract. The closing inference is hedged ('suggest') and grounded in the contig-boundary design. The limitations correctly note that the perturbation analysis is exploratory and that the benchmark scope is limited. The question follows from the exploratory analysis without asserting a conclusion.
An exploratory gene-input perturbation analysis further links representation sensitivity to experimentally measured fitness defects.
media — pass: The selected media is the verified direct arXiv abstract page for this paper, which is the most direct access to the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A model's initial training on a large dataset, before further adaptation for particular uses.
Definition referenceA list of numbers that represents a piece of text, an image or another item so software can compare meanings.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA model's initial training on a large dataset, before further adaptation for particular uses.
Definition referenceA list of numbers that represents a piece of text, an image or another item so software can compare meanings.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv