dimensionality reduction

Cortical geometry constrains the unimodal anchors of sensory integration

The paper tests if cortical geometry encodes sensory function by modeling multisensory integration from geometric eigenmodes, finding they align with sensory domains but require additional modes for full hierarchy.

Disentangling Brain-Psychopathology Associations: A Systematic Evaluation of Transdiagnostic Latent Factor Models

This study evaluates transdiagnostic latent factor models for brain-psychopathology associations using neuroimaging data, finding no consistent improvement in predictive utility over summary scores.

MIDFA: Scalable Bayesian Factor Analysis for Mixed and Incomplete Data

The paper introduces a scalable Bayesian factor analysis framework for handling mixed and incomplete data, demonstrating its effectiveness on MS datasets.

Positive Interpretation of Emotional Ambiguity Across Development: LC-dlPFC Circuitry

This study examines how positive interpretation of emotional ambiguity develops across age, showing it is linked to LC-dlPFC circuitry and evaluative drift bias rather than perceptual sensitivity.

Deciphering cross-omics complexity of tissues via diagonal integration of unpaired spatial multi-omics data

This paper introduces STAMO, a spatially aware integration method for unpaired spatial multi-omics data, showing improved alignment and domain detection compared to existing methods.

Compositionality of social gaze in the prefrontal-amygdala circuits

The paper identifies a compositional neural code for social gaze in macaques, showing how primitives like content and state are abstractly represented in prefrontal-amygdala circuits.

EEG microstate dynamics during psilocybin intoxication relate to acute experience and persisting psychological changes

This study examines EEG microstate dynamics during psilocybin intoxication and finds correlations between acute neural changes and long-term psychological effects.

Generalisable tissue-wide molecular reconstruction from histology

GHIST+ reconstructs tissue-wide molecular maps from H&E histology, enabling cohort-scale molecular reconstruction from sparse spatial transcriptomic data.

A quantitative coordinate system for developmental dynamics

This paper introduces STERN, a framework that quantitatively compares developmental trajectories across embryos by learning spatiotemporal representations from 4D imaging data.

TifBERT: a self-supervised foundation model for normalization-robust bulk RNA-seq representation learning

TifBERT is a self-supervised foundation model for bulk RNA-seq that learns normalization-robust representations without expression discretization or external embeddings, achieving high accuracy and capturing pathway-level biology.

JASMINE: A powerful representation learning method for enhanced analysis of incomplete multi-omics data

JASMINE improves analysis of incomplete multi-omics data by learning representations that preserve modality-specific and shared information, achieving better performance across tasks.

Robust metabolomics data normalization across scales and experimental designs

This paper introduces robust normalization methods for metabolomics data to reduce technical variance and improve downstream analysis.

BiXformer: A Bidirectional Cross Attention Transformer for Disentangling Inter-Regional Neural Dynamics

BiXformer disentangles bidirectional neural communication by modeling causal and acausal streams, revealing temporally structured components in neural dynamics.

Individualised mapping of living human brain mitochondria by MRI reveals signatures of bioenergetic defects.

This paper introduces a neuroimaging framework to map brain mitochondria non-invasively and finds age-related declines in mitochondrial density and respiratory capacity, along with disease-specific alterations and links to systemic stress and cognition.

FLIM Playground: An interactive, end-to-end graphical user interface for analyzing single-cell fluorescence lifetime data

FLIM Playground provides an interactive, open-source platform for analyzing single-cell fluorescence lifetime data, offering unified workflows and real-time analysis tools validated on cancer cell line datasets.

ConnectoFM: A Foundation Model for Learning the Language of the Connectome

ConnectoFM is a foundation model for connectomics that outperforms existing methods in low-data regimes and improves accuracy for critical biological structures.

FEABAS: A Stitching and Alignment Tool for Serial EM Data

FEABAS is a tool for efficiently aligning and stitching serial EM data, handling artifacts and enabling scalable vEM reconstruction.

A Unified Spatial AI Framework for Cross-Domain Tissue-State Analysis in Trauma, Oral, and Cardiovascular Pathology

This paper develops a spatial AI framework to identify conserved tissue-state organization across trauma, oral, and cardiovascular pathology using spatial transcriptomic data, finding six conserved fuzzy tissue states.

A New Hybrid Method for Brain Tumor Detection Based on Deep Learning

This paper introduces a hybrid DCNN-WOA framework for brain tumor detection and classification, achieving high accuracy and outperforming existing methods on an MRI dataset.

Sex-related structural alterations across common epilepsies: a worldwide ENIGMA study

This study examines sex differences in structural brain alterations in common epilepsies using neuroimaging data from a large international cohort, finding some regional differences but no significant sex-by-diagnosis interactions.

Hybrid Solid-Liquid Optics Enable Scalable, High-Resolution, Multi-Immersion Light-Sheet Microscopy

This paper introduces a new optical framework and imaging devices that enable scalable, high-resolution, multi-immersion light-sheet microscopy for biological samples.

The ENIGMA MEG Pipeline: Automated cortically localized spectral analysis of multi-site resting state MEG datasets

The ENIGMA MEG pipeline automates spectral analysis of multi-site resting state MEG data, demonstrating compatibility with large-scale processing.

Homophily-informed generative models of brain maps

This paper introduces a generative model of brain maps based on homophily and shows it can accurately reconstruct maps using biological and functional similarity.

CiliAI: Automated segmentation and compartment specific fluorescence quantification of primary cilia in confocal microscopy images

CiliAI automates cilia segmentation and fluorescence quantification in confocal images, showing agreement with manual methods and reducing analysis time.

Polynomial Trajectory Compression for Protein Language Model Embeddings

This paper introduces a method to compress protein language model embeddings using polynomial trajectories and PCA, achieving efficient storage and reconstruction without loss of downstream task performance.

Learning quality scores for chromatin accessibility bigWig tracks using Machine Learning

The paper develops a machine learning framework to assess quality of chromatin accessibility bigWig tracks using biologically relevant features and internal controls.

Re-evaluating the benefit of training scale in cell foundation models

This study evaluates how training scale affects cell foundation models by comparing performance with frozen representations and finds that larger pretraining data improves model performance.

Mapping Chemical Diversity: Descriptor-Guided Clustering of Natural Products in the COCONUT Database

The paper develops a framework to map natural product chemical space using descriptor-guided clustering in the COCONUT database, identifying optimal descriptors and clustering methods to characterize chemical diversity.

Compositional and interpretable representation of histology using AI foundation models and sparse autoencoders

This paper introduces a human-in-the-loop framework using AI foundation models and sparse autoencoders to interpret H&E histology images and integrate them with molecular spatial profiling.

Low-dimensional prefrontal representations of objects during working memory

The paper investigates the dimensionality of prefrontal cortex representations during working memory and finds they are low-dimensional, compressed compared to visual inputs.