The front page of AI-driven biology
Thursday, October 15, 2026Thu, Oct 15
Updated 7:01 PM PT
Investigators built predictive machine learning models to infer cell-type-specific gene expression from single-nucleus data generated in the multi-ancestry PsychAD cohort.
Researchers at MIT, Massachusetts General Hospital, and Harvard Medical School have developed RamanOmics, a noninvasive technique pairing Raman vibrational microscopy with single-cell gene expression to profile senescent cells.
DeepSCENIC transfers representations from sequence-to-function deep learning models, such as Enformer and Borzoi, to single-cell multiomic atlases to reconstruct causal gene regulatory networks.
A graph-guided autoencoder called MeRN inferred metabolic pathway activity from single-cell RNA sequencing data to pinpoint why folate deficiency disrupts mouse neural tube closure.
The Bayesian framework converts continuous methylation proportions into discrete states, factorizes the latent structure, and propagates measurement uncertainty into the final cell assignments.
A computational biology team developed scMGFGRN, a framework that predicts gene regulatory networks by fusing single-cell RNA sequencing data with DNA sequence embeddings and Gene Ontology hierarchies.
ConvexGating uses machine learning to convert high-dimensional single-cell clusters into interpretable, low-contamination gating strategies that hardware cell sorters can execute directly.
Computational biologists developed SPHERE-PCA, a method that maps the first three principal components of single-cell RNA sequencing data onto a three-dimensional unit sphere to create an interpretable coordinate framework.
A spatial multi-omics study tracking over 112 million cells across three melanoma neoadjuvant immunotherapy regimens revealed that mature tertiary lymphoid structures with germinal centers, sustained by specialized fibroblast scaffolds, distinguish responders from non-responders.
A computational method named MESIC uses a biomedical language model to translate written gene annotations into low-dimensional semantic coordinates that give biological meaning directly to single-cell data.
A computational method called ssJSD groups single-cell chromatin contact maps by distinguishing biologically meaningful uncontacted DNA regions from technical dropouts caused by shallow sequencing.
Researchers have developed WAVE, a neural network designed to predict transcriptional changes caused by small molecules across both bulk tissue and individual cells.
The immgenT consortium developed immgenT-GP, a computational framework that deconvolves roughly 800,000 mouse T cells into 200 distinct, reproducible gene expression programs.
High glucose levels induce direct tissue-level inflammation and epithelial cell detachment in human stem cell-derived kidney organoids, according to a study led by researchers at the University of Washington and the University of Miami.
A computational team introduced AmbientMapper, a generative model designed to decontaminate high-throughput single-cell combinatorial indexing experiments by learning ambient background profiles from empty droplets.
A machine-learning algorithm called PreGame identifies tumor-reactive gamma delta T cells from single-cell surface-protein and RNA sequencing data (CITE-seq), separating functional killers from bystander immune cells.
Researchers built IGNITE, a semi-supervised machine learning framework that ranks therapeutic targets by combining human genetic evidence with Perturb-seq profiles from 22 million primary human CD4+ T cells.
A multi-agent framework called scACORN coordinates specialized small language models rather than a single monolithic model to interpret single-cell transcriptomic profiles.
Researchers built REN-former by fine-tuning the Geneformer foundation model on single-cell RNA sequencing from 45 kidney patient samples to predict directional cell shifts from acute kidney injury to chronic kidney disease.
A computational study showed that splitting single-cell datasets randomly between training and test sets leaks labels through related clonal sisters, patient backgrounds, or experimental batches.
LucaCell is a single-cell foundation model that encodes transcriptomic features through pre-trained mRNA nucleotide sequence embeddings rather than fixed gene names, allowing it to translate across species without manual ortholog mapping.
Therna Biosciences has released public datasets from its Chronos platform profiling roughly 60,000 synthetic regulatory elements across approximately 50 pooled cell lines.
Researchers introduced ICARus, a computational framework that uses protein language model predictions to reduce off-target noise in RNA interference functional screens.
A deep neural network evaluated 274,959 clinical lentiviral integration sites across 10 patients treated for severe combined immunodeficiency (SCID-X1) to examine where vectors insert into human DNA.
Computational biologists introduced PHAROS, an algorithmic framework that chains predictions from existing single-cell perturbation models to simulate and search for synergistic two-drug combinations.
A systematic evaluation of 13 computational methods across 25 perturbation datasets found that current models struggle to generalize cellular drug and gene responses across different cell types.
DiffusionOT applies neural networks and optimal transport to model how cells differentiate over time by accounting for random variations in gene expression.
An independent benchmark of 16 computational models found that methods designed to predict single-cell gene expression directly from standard tissue microscopy images perform poorly across skin disease samples.
scFLAME is a generative probabilistic model that performs dimensionality reduction, batch correction, cell clustering, and marker-gene identification in a single step from single-cell RNA sequencing counts.
CellART is an open computational method that segments individual cells and identifies their cell types directly from high-resolution spatial transcriptomics data, which measures gene activity across tissue slices.