benchmark study

Probing Genomic Foundation Models with Splice-Variant Perturbations

This paper evaluates genomic foundation models using splice-variant perturbations, finding they match task-specific networks and generalize across mutation types.

Comparative performance of reference-based metagenomic tools to identify species-level taxa among families of bacteria: benchmarking Mycobacteriaceae and Neisseriaceae

This study benchmarks reference-based metagenomic tools for species-level detection in closely related bacterial families, finding metapresence and sylph perform best.

Illuminating dark genes and dark metabolites: calibrated functional association across kingdoms with autoencoders and correlation

VaLPAS pairs a bi-directional autoencoder with correlation for omics guilt-by-association with calibrated confidence; in L. starkeyi its autoencoder linked 176 unidentified metabolites to known-function proteins at ≥50% confidence.

SpatialOmicsLab: an integrated research environment for AI co-scientists in spatial transcriptomics

SpatialOmicsLab provides an integrated environment with AI tools for spatial transcriptomics analysis, outperforming expert workflows in multiple tasks.

Profiling peripheral immune cells in Parkinsons disease: A Scoping Review

This review synthesizes findings on peripheral immune cell changes in Parkinson's disease, identifying common immune signatures and methodological inconsistencies across studies.

Reliable extraction of clinical-trial evidence with provenance, verification, and targeted review

The paper evaluates EviSearch, a clinical evidence-curation system, showing high accuracy and improved results with targeted review.

Additive-input encoding fails operator-level target-held-out prediction on current Perturb-seq screens

The paper shows that additive-input models fail to generalize in Perturb-seq screens when predicting new perturbation targets.

Continuous Integration and Software Quality in Scientific Software: A Large-Scale Empirical Analysis of GitHub Repositories

This study analyzes GitHub repositories to show that continuous integration and testing improve software quality and academic impact in scientific software.

Characterizing and Mitigating Protocol-Dependent Gene Expression Bias in 3' and 5' Single-Cell RNA Sequencing

This study identifies and mitigates protocol-dependent gene expression bias between 3' and 5' scRNA-seq by focusing on a small set of biased genes rather than applying broad normalization.

Careers in Scientific Editing: Results from the 2024 Scientific Editors Network (ScENe) Survey

This study presents findings from a 2024 survey of 132 scientific editors, revealing demographic homogeneity and variability in career paths, with implications for integrating scientific editing roles into research development.

Trait-Based Selection of Annual Plant Species for Enhancing Ecosystem Functioning in Extensive Green Roof Systems: A Systematic Review

This review identifies annual plant species suitable for extensive green roofs and highlights functional traits important for their success in rooftop conditions.

A global genomic survey of prokaryotic carbon fixation reveals an oxygen-tolerant rTCA cycle in the surface ocean

The paper surveys global prokaryotic carbon fixation pathways, revealing an oxygen-tolerant rTCA cycle in surface ocean Campylobacterota lineages.

Informatics Methods to Enhance Bacterial Peptide Detection

The paper develops and benchmarks an in silico spectral library workflow to detect low-abundance bacterial peptides in human-dominated samples, finding it has complementary performance to conventional database searches.

Learning source weights for target-position estimation: comparing fixed weights and an input-dependent rule

In simulation, under reduced accuracy or temporally correlated errors, recalibration or fixed-weight learning captured most gains; a disagreement-dependent rule beat fixed weights mainly under occasional large errors (4.05% lower MSE).

A decision-based criterion and a resolution threshold for ranking genomic prediction models within trials

The paper introduces a decision-based criterion and resolution threshold to evaluate genomic prediction models by analyzing their ranking stability and accuracy differences across trials.

MetagenomicsBench: A Verifiable Benchmark for Agentic Metagenomics Analysis

MetagenomicsBench evaluates AI agents' ability to analyze metagenomic data through 100 verifiable tasks, finding that performance varies across domains and that technical plausibility does not guarantee correct biological interpretation.

Joint protein, mRNA and DNA sequence design and optimization with nucleotide-level Potts models

A structure-conditioned nucleotide-level Potts model designs protein and coding sequences jointly, enabling tRNA adaptation and RNA motif embedding beyond synonymous codons with designability comparable to Caliby.

Characterization of cross-family conserved fragments

The paper characterizes cross-family conserved fragments, showing their structural and biophysical properties and their role in folding and stability.

WrapEvoFS enables auditable feature compression with regret-constrained representative locking

WrapEvoFS selects a representative feature set from stochastic searches with regret constraints, ensuring auditable and reproducible results without claiming predictive superiority.

Phenotypes of ultra-rare variant carriers benchmark variant effect scores

The paper benchmarks variant effect scores using UK Biobank data on ultra-rare variant carriers, showing coding and splicing scores better predict protein abundance and traits than transcriptional regulation models.

Complete pangenome graph annotation with GrAnnoT and GrAPEx

The paper introduces GrAnnoT and GrAPEx to fully annotate pangenome graphs, integrating annotations from multiple genomes and annotating non-annotated regions, reducing reference bias.

Reliable Evaluation of Transductive Population Graph Neural Networks for Multisite Autism fMRI

Site-aware population GNN on ABIDE-I: cohort AUC about 0.94 (kept with site-only edges) was not reproduced by a site-prior classifier (0.506); a directed-edge support gain (+0.252) vanished when symmetrised (−0.003).

GroundAnnot: a closed-vocabulary contract for grounding LLM gene-set annotation in live enrichment backends

GroundAnnot ensures LLM gene-set annotations are grounded in real enrichment results from backends like PANTHER and Enrichr, improving accuracy by restricting outputs to validated GO terms.

Dataset structure outweighs method choice in single-cell cell-type annotation

In a 63-tool benchmark, dataset identity explained ≈84% of Cohen's κ variance on synthetic data versus ≈4% for tool identity, and k-nearest-neighbor (kNN) purity predicted accuracy near-linearly (R² = 0.85–0.99).

Rank Normalization Enables Cross Pathway Comparison in Degradation Resolved Proteome Turnover Analysis

The paper introduces rank normalization to compare protein turnover across degradation pathways, revealing pathway-specific substrate preferences and proteaphagy.

Accurate ab initio gene prediction in eukaryotes with Tiberius in multiple clades

Extended Tiberius models outscore Helixer and ANNEVO on all 33 genomes from the five clades all three tools cover, and approach BRAKER3 accuracy in four clades without extrinsic evidence.

Octave: Scale-resolved Evaluation of Spatial Gene Expression Prediction from Histology

The paper introduces OCTAVE, a scale-resolved evaluation method for spatial gene expression prediction from histology, showing that fine-scale differences are missed by standard metrics.

Registered Report: Replication and Extension of Nozaradan, Peretz, Missal and Mouraux (2011)

This study replicates Nozaradan et al.'s 2011 findings on neural activity during imagined beat perception, finding smaller effect sizes and no reliable replication of main effects.

When to Trust Epigenetic Clocks: Avoiding False Positives in Aging Interventions.

This study evaluates epigenetic clocks for aging interventions, finding newer clocks trained on mortality or aging rates are more reliable than those based on chronological age.

Pretrained gene representations transfer mean expression more broadly than spatial patterns in virtual spatial transcriptomics

This study evaluates how pretrained gene representations contribute to virtual spatial transcriptomics by separating their effects on mean expression and spatial patterns across multiple datasets.