Single-cell droplet method tracks blood stem cell clones using DNA methylation
Researchers developed Droplet-MT, a microfluidic method that profiles the transcriptome and DNA methylation together across roughly 10,000 single cells per run.
Preprints
Recent original research from bioRxiv and arXiv. Preprints have not been peer reviewed.
Page 3
Computational design tailors light-driven excited states in custom-built proteins
Researchers paired computational protein design with quantum mechanics and molecular dynamics simulations to program photochemical behavior in synthetic proteins built from scratch.
Vision transformer identifies bacterial species from label-free microscope videos
Researchers adapted the DINOv3 vision transformer to classify five common bacterial pathogens from phase-contrast time-lapse imaging of cells grown in microfluidic traps without fluorescent dyes.
Deep learning ranks cancer drugs within individual tumors rather than predicting broad sensitivity
Researchers developed DEPICT, a deep learning model trained on 231,880 cell line-drug pairs across 941 cancer cell lines and 289 compounds to rank which therapies work best inside a specific patient tumor profile.
Interactive AI pipeline speeds up live-cell tracking by learning from human edits
A new cell-tracking framework uses human-in-the-loop feedback to propagate manual corrections across an entire live-microscopy dataset rather than editing one track at a time.
Computational screen pinpoints mutations that shape tau into brain trauma fibrils
A screen of approximately 12,000 mutations across six tau fibril structures identified key residues that bias misfolded tau proteins toward folds seen in chronic traumatic encephalopathy instead of Alzheimer's disease.
Long-read sequencing links single-molecule chromatin packing to alternative gene transcripts
A computational pipeline called EpiFLAIR integrates long-read Oxford Nanopore sequencing of chromatin fibers, nascent RNA, and mature transcripts in yeast.
Generative model designs proteins around unnatural amino acids and chemical warheads
A generative framework named CheMoDesign designs proteins from scratch around user-selected noncanonical amino acids, post-translational modifications, and synthetic reactive groups.
Image-based machine learning flags four repurposed drugs that block dengue infection
Investigators paired high-content microscopy with a machine learning pseudotime model tracking dengue virus proteins over time to screen 960 repurposed compounds in infected cells.
Spatial atlas maps lung tumor invasion and exposes technical traps in tissue depth
Researchers combined single-nucleus RNA sequencing across 413,697 nuclei with spatial transcriptomics in 56 lung sections to trace the progression of lung adenocarcinoma from early lesions to invasive cancer.
Specialized splicing models spot hidden RNA errors behind neurodegeneration better than general genomic AI
Researchers compared the specialized splicing model OpenSpliceAI against the genomic foundation model AlphaGenome to predict rare, disease-linked RNA splicing errors called cryptic exons caused by TDP-43 dysfunction.
Vision language models interpret multiplexed cell imaging screens
Genentech researchers developed CellMAPP and an analytical framework called CHART to track organelle changes across 524 genetic knockdowns in breast cancer cells.
Engineered polymerase reads modified synthetic RNA
Researchers evolved a heat-stable DNA polymerase from the bacterium Thermus thermophilus into a reverse transcriptase that can read and copy C2'-modified synthetic RNA.
Generative model designs culture conditions to steer human blood stem cells
Researchers built LabCompass, an active-learning system that pairs a flow-matching generative model with a surrogate predictor to design continuous chemical recipes that push stem cells toward specific fates.
Screening and structural modeling flag the approved drug aprepitant for a rare brain disorder
Researchers combined a phenotypic screen of around 1,200 approved drugs with Boltz-2 structural modeling to identify inhibitors of NALCN, a sodium leak channel mutated in the ultra-rare neurodevelopmental condition CLIFAHDD.
Physical cell crowding guides reinforcement learning for adaptive cancer therapy
Researchers combined two-dimensional melanoma cell cultures with mechanical particle simulations to study how spatial crowding drives drug-resistant tumor cells to break free.
Designed mini-protein blocks brain ion channel and reduces stroke damage in mice
Researchers used deep-learning design methods to build Denasin1, a subtype-selective mini-protein inhibitor targeting the human acid-sensing ion channel hASIC1a.
Language model predicts molecular dynamics trajectories across diverse proteins
Researchers introduced MD-LLM-2, a transferable language model that generates protein motion trajectories conditioned on sequence, temperature, and structural history.
Hybrid neural model predicts human drug exposure across unfamiliar chemical structures
Researchers developed PK-MUSE, a computational model that marries traditional two-compartment physiological equations with neural networks to predict human drug concentration over time directly from chemical structure.
AI agent builds spatial transcriptomics pipelines using a standardized software suite
Researchers built SpatialOmicsLab, an evaluation suite combining 113 analytical tools, 63 reference datasets, and 4,614 task specifications for spatial transcriptomics, alongside an AI agent named ST-Coscientist designed to assemble and run those workflows.
Structural features outperform protein language models on shape-shifting proteins
Computational biologists evaluating metamorphic proteins, which switch between distinct three-dimensional folds, found that protein language models rely heavily on sequence memorization and fail to generalize to novel shape-shifting regions.
Deep learning maps how easily flu strains jump between animals and humans
Researchers built a deep-learning model called InfluProto that evaluates the cross-species transmission risk of influenza A by framing host adaptation on a continuous spectrum rather than as rigid categories.
Computational protein design steers sugar structures on custom glycoproteins
Researchers built an all-atom computational design method that controls the composition of attached sugars, known as N-linked glycans, by shaping the surrounding protein backbone.
Targeted DNA model recovers novel enzymes from metagenomes at lower computing cost
Researchers built REFinder, a pipeline that uses a DNA language model called REBEAN to label raw sequencing reads by enzyme class and assemble only sequences with putative catalytic signatures.
Vision-language model reads histology images and spatial gene data together
Researchers developed VINTER, a generative vision-language model trained on 1.45 million histology-caption pairs that jointly processes tissue morphology, local gene counts, and spatial neighborhoods.
Generative model designs synthesizable drugs from an 11-trillion molecule library
Researchers built Hyper Screening X, a generative flow network that bakes reaction rules directly into design algorithms to search an 11-trillion make-on-demand compound library while scoring only 10 million candidates.
Neural networks read electrical signals from nanopore protein sequencing
Researchers built an automated computational pipeline to segment and classify the ionic current blips produced as individual peptides pass through a nanopore.
Transfer matrices unify and accelerate RNA secondary structure calculations
Researchers replaced decades-old loop-specific nearest neighbor rules for nucleic acid folding with a single transfer matrix formulation that models the free energy of all loop types as the trace of a matrix product.
Deep learning assesses cell viability from standard microscope images without dyes
Researchers built ViabiLens, a computer vision system that couples a cell detection algorithm with a convolutional neural network to predict whether cells are alive or dead using unstained brightfield microscopy.
Desktop app uses neural networks to speed up mass-spectrometry lipid separation
Researchers released a standalone desktop application for intelligent Differential Mobility Spectrometry, a deep learning tool that automates parameter tuning for distinguishing nearly identical lipid molecules.