M Kellis

This study integrates multi-omic data from mouse and human choroid plexus to identify conserved molecular programs of epithelial function across species.

ProxiNet transfers spatially learned cellular proximity from spatial transcriptomics to dissociated single-cell transcriptomes, revealing tissue organization and age-related changes in Alzheimer's.

The paper identifies 1,596 Consensus HAQERs by aligning T2T assemblies, showing they are rapidly evolved regions with mutation signatures, regulatory functions, and disease associations.

GeneSIS improves polygenic score transferability by incorporating gene-by-sex interactions, enhancing prediction accuracy in non-European populations for 32 traits.

This study uses sequence-to-function deep learning to identify lineage-specific cis-regulatory elements in human evolution, finding they are shorter, less conserved, and enriched in neurodevelopmental pathways.

The paper introduces HI-JEPA, a model trained on molecular proximity measurements to predict protein interactions and disease-related configurations.

This paper develops an extended polygenic risk score (ePRS) framework for type 2 diabetes that incorporates gene-gene and gene-environment interactions, showing improved predictive power over traditional additive models.

ProtJEPA predicts joint embeddings across ten biological modalities from protein sequences, outperforming baselines in zero-shot GO retrieval and enzyme class prediction.

This paper introduces Phenoverse, a deep learning framework that learns disease state representations from single-cell transcriptomic data, enabling disease prediction and severity assessment using binary labels.

The paper identifies 1,400 stop codon readthrough genes in Drosophila, reveals sequence features associated with programmed readthrough, and shows conserved cis-elements functional in mammalian cells.

The paper identifies circadian misalignment between cancer and immune cells in breast tumors as a mechanism of immune evasion and therapeutic resistance.

The paper introduces ExRPath and 15ExMPath techniques to expand human brain tissue for nanostructure analysis, revealing amyloid nanoclusters associated with SARS-CoV-2 in COVID-19 patients.

This study uses polygenic scores to identify genetic associations with Alzheimer's disease phenotypes and subtypes across multiple cohorts.

This study identifies length-dependent hypercontractility as the initial phenotype of MYBPC3-haploinsufficiency in iPSC-derived cardiac tissues, decoupled from early transcriptional and metabolic changes.

The paper identifies 25% of 7,264 non-canonical open reading frames (ncORFs) as producing translated gene products, supported by over 3,000 peptides from 3.8 billion mass spectra.

The paper identifies a vulnerable astrocyte state in the prefrontal cortex of Down syndrome-associated Alzheimer's disease through integrated single-nucleus RNA-seq and ATAC-seq analysis.

Hydra is a deep generative framework for single-cell omics data that improves cell type annotation and feature selection, particularly for rare populations.

The paper develops BayesMonSTR to detect mosaic STR mutations at single-cell resolution and finds increased STR indels in aging cells, particularly in prefrontal cortex neurons.

This study shows that Scn4b modulates Huntington's disease severity by affecting gene expression and neuronal function in mouse models.

This paper identifies asymmetric-drive rectification as a mechanism converting stress into structural memory, enabling functional specialization in cells and adaptive computational systems.

The paper identifies convergent signals that encode epigenetic memory of adaptive states in NK cells under dietary restriction, showing how these signals reduce DNA damage and enhance metabolic plasticity.

The paper reconstructs metabolic networks across 58 human tissues and cell types using RNA-seq and epigenomic data, identifying core and cell-type-specific metabolic functions and their regulatory elements.

The paper identifies metabolic human accelerated regions (mHARs) in adipose tissue that coordinate adaptive responses to nutrient deficit, revealing their role in maintaining metabolic homeostasis and resilience.

sc4D integrates optimal transport with autoencoder embeddings to analyze spatio-temporal single-cell transcriptomics data, revealing glial responses to Alzheimer's disease pathology and predicting therapeutic interventions.

PRISM improves polygenic score transferability across populations by integrating ancestry-aware tissue-specific genomic annotations, achieving a 13.10% performance boost for African ancestry.

CONCERT, a niche-aware generative model, predicts spatial perturbation responses more accurately than existing methods across multiple tissues and experimental setups.

The paper infers enhancer activity from cfDNA fragmentation patterns and proposes a chromatin fiber model to explain these signatures.

The paper constructs a 3D human brain model integrating multiple cell types to study Alzheimer's Disease mechanisms linked to APOE4.

This study examines how exercise training affects subcutaneous white adipose tissue transcriptomes in individuals with obesity and T2D, finding that exercise reverses detrimental gene expression patterns and correlates with improved metabolic markers.

The paper identifies a mature dendritic cell subtype associated with improved immune response and prognosis in melanoma patients treated with anti-PD-1 checkpoint inhibitors.