benchmark study

Benchmarking AI-generated structural ensembles of membrane proteins against physics-based modelling

This study benchmarks AI-generated structural ensembles of membrane proteins against physics-based modeling, showing BioEmu can generate plausible conformations for large membrane proteins with lower computational cost.

A Framework for Benchmarking Pathway Reconstruction Algorithms

This paper benchmarks 14 pathway reconstruction algorithms across 822 datasets using a standardized framework to evaluate performance, similarity, and computational efficiency.

A Systematic Evaluation of Single-Cell Batch Integration Metrics and sBEE, A New Metric

The paper evaluates existing batch integration metrics in scRNA-seq and introduces sBEE, a new metric that provides more reliable assessments of batch mixing across diverse scenarios.

FLIP2: Expanding Protein Fitness Landscape Benchmarks for Real-World Machine Learning Applications

FLIP2 adds seven protein fitness datasets and 16 splits for enzymes, light-sensitive proteins, and interactions; simple ridge and zero-shot baselines often rival fine-tuned protein language models.

Benchmarking single-cell foundation models in a zero-shot setting

This study benchmarks four single-cell foundation models across four tasks, finding they do not consistently outperform traditional methods but show promise in protein expression prediction.

FIDDL: depth-matched negative controls distinguish genuine interspecific introgression from competitive-mapping artifact

The paper introduces FIDDL, a tool to distinguish genuine interspecific introgression from artifacts by using depth-matched negative controls and locus-level tests.

AdaptivePy: a unified Python framework for adaptive sampling in molecular dynamics

AdaptivePy provides a unified framework for adaptive sampling in MD simulations, implementing nine seed-selection policies and enabling meta adaptive sampling schemes.

MethylBench: A comprehensive benchmark of DNA methylation profiling methods across diverse sequencing platforms

MethylBench benchmarks DNA methylation profiling methods across sequencing platforms, finding consistent DMC hotspots and highlighting platform-specific strengths and limitations.

Detecting pathogenic structural variation in families with undiagnosed rare disease in a national genome project

This study uses long-read sequencing to identify pathogenic structural variants in undiagnosed rare disease families, increasing diagnostic yield in three families.

A global database of insect traits and anthropogenic associations.

The paper introduces AnthropInsect, a database of insect traits and human associations, containing 5867 species across six orders, to support ecological and conservation research.

Where are they now? Academic and career trajectories of national laboratory STEM internship alumni from community colleges, compared to those from universities

This study compares the academic and career outcomes of community college and university students who participated in DOE national lab internships, finding comparable STEM success rates between the groups.

A benchmarking framework for single-cell genome-scale metabolic model construction

This study benchmarks strategies for constructing single-cell genome-scale metabolic models across nine scRNA-seq datasets, evaluating accuracy, sensitivity, and computational feasibility.

The Fitness Cost of Therapeutic Resistance in Cancer: A Systematic Review

This paper reviews 49 experiments showing that therapy-resistant cancer clones often have a fitness cost, with two-thirds demonstrating competitive disadvantage in untreated environments.

Benchmarking Generalizability in Deep Learning-Based White Matter Tract Segmentation

This study benchmarks deep learning models for white matter tract segmentation across diverse datasets, finding that CNN-based models perform best within their training domain but degrade under domain shifts, and introduces an ensemble method to mitigate this issue.

Area under the curve quantification outperforms spectral counting in metaproteomics, but matching between runs is detrimental

The paper compares AUC and SpC quantification methods in metaproteomics, finding AUC outperforms SpC in accuracy and dynamic range, but MBR harms AUC results.

SVkhor: a unified framework for structural variant integration across long-read, short-read, and optical genome mapping data

SVkhor normalizes, merges and integrates long-read, short-read and optical-mapping structural variant calls; on HG002 it retained 98.6% of naive concatenation's true positives with 58.2% fewer false positives.

Benchmarking Deep Learning Predictions of Mutation-Induced Fold Switching

This study benchmarks deep learning models' ability to predict mutation-induced fold switching in proteins using an NMR-characterized dataset of fold-switching mutants.

A complete diploid human genome benchmark for personalized genomics

T2T-HG002v1.1, a complete diploid HG002 assembly lacking detectable errors over 99.4% of its length, enables genome-based benchmarking of assemblies, sequencing reads, and phased variant calls.

Modeling the structure-conditioned sequence landscape for large-scale protein design with TriFlow

TriFlow, discrete flow-matching sequence designer with global three-track structure attention, raises in silico binder success over ProteinMPNN in RFdiffusion and BindCraft-style pipelines on 500 targets and flags functional sites.

From sound to source: Human and model recognition of environmental sounds

Neural networks trained to detect sound categories in synthetic multi-source scenes approach human accuracy on the EnvAudioEval behavioral benchmark and track human patterns better than cochlear or cortical filterbank baselines.

PlumageParts: A fine-grained avian plumage segmentation dataset and benchmark for ecological image analysis

PlumageParts provides a dataset of 4,705 bird images annotated for nine plumage regions and benchmarks segmentation models, achieving 84.01% mean Intersection over Union with a frozen DINOv3 encoder.

Structured Schemas for Provenance-Rich, LLM-Assisted QSP Model Calibration

The paper introduces MAPLE, a framework that uses structured schemas to improve LLM-assisted QSP model calibration by ensuring data provenance and reducing hallucination.

AI4Life Open Calls and Public Challenges: why, how, and what we have learned.

The paper discusses AI4Life's Open Calls and Public Challenges, highlighting the state of bioimage analysis, tool availability, and the need for better data and infrastructure.

Spatial profiling identifies a GLIS2-associated fibroblast state linked to peritoneal recurrence in gastric cancer

Pan-cancer benchmarking reveals complementary copy number signatures with distinct multi-omic predictability

This study benchmarks three copy number signature compendia across cancers, showing their distinct predictability from multi-omic data and highlighting framework-dependent performance.

Epigenetic and brain age across development: Performance and associations in the MIND consortium

Across 15 cohorts (birth to 24 years), 20 epigenetic clocks and eight brain age models predicted chronological age modestly, stabilizing by middle childhood; their age residuals were weakly associated.

How Bias Shapes the Leaderboard: Scoring Function Performance Under Scrutiny

This study evaluates how systemic biases affect scoring function performance in virtual screening, finding that biases can inflate results and that classical methods outperform machine learning models on out-of-distribution targets.

BioPathfinder: Evidence-guided multi-agent platform enables hypothesis discovery for CAR-T engineering

A provenance-tracked multi-agent system curated 57 CAR-T patient scRNA-seq studies, generated reviewable hypotheses, and prioritized KLRC1/NKG2A, whose blockade improved CD8+ CAR-T antitumour activity in a Raji xenograft model.

Comparison of multiple video tracking-based behavioral summary approaches for compound discrimination

Scalar keypoint aggregates matched Keypoint-MoSeq, VAME, SimBA and A-SOiD at discriminating eight drug-dose treatments across 314 mouse open-field recordings; social context did not improve recall.

OmniAge: a compendium of aging omic biomarkers links mitotic clocks to clonal hematopoiesis and causality

OmniAge compiles 413 aging biomarkers across 12 categories and reveals associations between mitotic clocks, clonal hematopoiesis, and cancer risk using blood datasets.