benchmark study

Classical baselines outperform released deep-learning ITS classifiers, which collapse on ITS2 where predictions follow the flanking regions

MycoAI ITS classifiers trail alignment, SINTAX and HiTaC on full-length ITS and drop to 28.3% and 18.5% seen-genus family accuracy on ITS2, as predictions track flanking regions.

Spurious model comparisons are widespread in biomedical artificial intelligence

97% of 184 high-impact biomedical AI studies compare cross-validated models with tests assuming fold independence; in 69% of re-analyzable studies a significant comparison does not survive correction.

Benchmarking long-read RNA sequencing for de novo transcriptome assembly in non-model plant species: insights from Moricandia arvensis

The paper benchmarks long-read RNA sequencing pipelines for de novo transcriptome assembly in non-model plants, finding organ-specific optimal strategies for flower and leaf tissues.

BenchHub enables an inclusive and transparent ecosystem for community-focused benchmarking in computational biology

BenchHub provides a modular ecosystem for community-focused benchmarking in computational biology by linking datasets, metrics, and ground truth to improve reproducibility and sustainability.

Genetic architecture of lipoma susceptibility implicates telomere biology

This study identifies genetic variants associated with lipoma susceptibility and links them to telomere biology, showing a positive association between genetically predicted longer telomeres and lipoma risk.

A Systematic Comparison of Single-Cell Perturbation Response Prediction Models

The paper benchmarks 13 models for predicting single-cell perturbation responses across 25 datasets, finding performance varies with perturbation effect size and evaluation metric.

Predicted and experimental protein-ligand coordinates with distance labels and evaluation splits

The paper introduces PLI-Parallax, a resource that combines predicted and experimental protein-ligand coordinates with evaluation splits and accuracy estimates for 30,567 systems.

Building an open infrastructure for molecular neuroimaging: standards and tools from the OpenNeuroPET initiative

The paper describes the OpenNeuroPET initiative, which develops standards and open-source tools to standardize and share PET neuroimaging data, enhancing reproducibility and collaboration.

Specification dependence of EEG theta/alpha ratio associations with cognitive impairment: a multiverse analysis

The paper evaluates the robustness of theta/alpha ratio as a cognitive impairment marker using multiverse analysis across multiple EEG preprocessing specifications.

A shared functional organisation underlies vascular disease remodelling

The paper identifies a reproducible functional organisation in vascular remodelling across diseases, showing coordinated rebalancing of biological functions rather than isolated pathway changes.

Thoracic Aortic Disease in Genetic Syndromes Beyond Established Aortopathy Genes

The paper identifies de novo pathogenic variants in syndromic genes associated with thoracic aortic disease beyond established aortopathy genes.

A metadata managed FAIR end-to-end workflow for microbial community Omics data analysis

The paper presents a metadata-managed FAIR end-to-end workflow for microbial community Omics data analysis, demonstrating its reliability and reproducibility through mock datasets.

BioSecBench-Function: A Verifiable Benchmark for Reasoning about Biological Function from Experimental Data

The paper introduces BioSecBench-Function, a benchmark for assessing AI agents' ability to infer biological function from experimental data, showing varying performance across models and tasks.

The effect of psychedelics on associative learning: a systematic review

This systematic review examines the effects of psychedelics on associative learning in humans and animals, finding enhanced learning in animals but inconsistent results and methodological heterogeneity.

Systematic Comparison of 10X Genomics and Parse Single Cell RNA Technologies across PBMC and CD8+ TEMRA Cells

Side-by-side PBMC and CD8+ TEMRA profiling shows Parse favors longer transcripts but underdetects cytotoxic genes (GNLY, PRF1, GZMB), so 10X Chromium better resolves TEMRA subsets.

A Comparative Evaluation of Structural MRI Foundation Models for Age, Sex, and Body-Mass Index Predictions

This study benchmarks four structural MRI foundation models for age, sex, and BMI prediction, finding that 3D-Neuro-SimCLR and AnatCL outperformed baselines overall.

Systematic benchmarking of commercial workflows for isoform-resolved single-nucleus transcriptomics

This study benchmarks four commercial LR snRNA-seq workflows, finding 10x Genomics 3' and 5' to be the most balanced for isoform-resolved single-nucleus transcriptomics.

Hierarchical Breakdown of RNA Structure Prediction in CASP16: From Reliable Local Helices to Speculative Multimer Assembly

This paper analyzes RNA structure prediction challenges in CASP16, showing that accuracy decreases from local helices to multimer assembly due to monomer errors and interface modeling issues.

Stoichiometry-dependent specificity in biotin enrichment: a benchmarking framework for proximity labeling proteomics

The paper benchmarks biotin enrichment in proximity labeling proteomics, showing that higher biotinylation improves specificity while lower levels increase background noise.

Cross-domain confidence reliability and remappability of frozen single-cell representations

The paper evaluates the reliability of confidence measures from single-cell models on new data and shows that raw confidence scores are unreliable, requiring recalibration with target-specific labels.

Exploring the feasibility, acceptability and impact of genomic newborn screening for rare diseases in England: A study protocol for the Generation Study - Process and Impact Evaluation

A configuration-resolved benchmark of differential abundance analysis methods for human gut 16S rRNA microbiome data

The paper benchmarks differential abundance analysis tools for human gut 16S rRNA data across configurations, showing that settings significantly impact performance and must be reported for interpretability.

Non-Covalent Poly(ADP-ribose) Signaling Organizes a Circadian E3 Ligase Network in the Brain

The paper identifies circadian-organized non-covalent PAR-binding E3 ligases in the brain, linking PAR signaling to time-dependent ubiquitin control and disease-relevant biology.

Long-Term Impact of Cumulative Hyperglycaemia on DNA Methylation and its Role in Diabetic Kidney Disease

The paper identifies DNA methylation changes linked to cumulative hyperglycaemia in type 1 diabetes and links some to diabetic kidney disease through Mendelian randomization.

Accessible and reproducible deployment reveals the practical boundaries of single-cell foundation models

Across 13 single-cell foundation models and 94 datasets, PCA and established methods usually match or beat them; pretrained embeddings help mainly with scarce labels, rare cells and open-set detection.

CROWN: Curated Repository Of Well-resolved Noncovalent interactions

An automated pipeline turns the PDB into 178,263 repaired, protonated, restraint-minimized protein-ligand complexes; scoring functions trained on them beat HiQBind-trained ones on CASF-2016 ranking and docking power.

The first OpenBind release: An open experimental structure-affinity dataset and benchmark for structure-based AI

A public crystallographic structure and affinity dataset for enteroviral 2A protease (925 binding events, 699 compounds), with docking, cofolding and screening baselines. Fragment fine-tuning of OpenFold3-p2 improves follow-on predictions.

PhageTAILor leverages machine learning for phage tail-like elements detection and classification in plant-associated bacteria

PhageTAILor uses LightGBM on 74 detector features to flag prophages and bacterial phage tail-like elements (tailocins, T6SS, eCIS), recovering 12 of 13 validated tailocins under leave-one-genus-out.

TigerAI: An AI-powered genetic evidence platform to support clinical development

Domain-grounded prompting lets GPT-5 produce genetic evidence for 13,022 drug target-indication pairs that predicts clinical success as well as expert curation, with wider coverage.

Audited vibe coding suggests partial fetal-like convergence of tumor proteomes

The paper reanalyzes a proteomic atlas using audited vibe coding to assess if tumors show fetal-like proteomic convergence, finding partial alignment with reverse fetal-to-adult maturation.