The front page of AI-driven biology
Thursday, October 15, 2026Thu, Oct 15
Updated 9:01 PM PT
Recent original research from bioRxiv and arXiv. Preprints have not been peer reviewed.
Researchers combined high-content imaging of patient-derived human liver organoids with machine learning to identify drugs that cause liver injury, classifying 54 known hepatotoxicants and 20 non-toxic controls.
A computational benchmark evaluated ten large language models on linking 5,861 compounds from the ChEMBL database to 14 adverse outcome pathways spanning five neurodevelopmental toxicities.
Researchers developed G4SCOPE, a convolutional and transformer neural network that predicts the chromatin occupancy of four-stranded DNA structures known as G-quadruplexes straight from primary sequence.
Machine-learning analysis of temporal changes across the cell-surface proteome identified the enzyme ALPP as an adaptive target that pancreatic tumors produce in response to KRAS inhibition.
Researchers introduced AuditPPI, an interpretability framework that converts sparse-autoencoder representations from frozen protein language models into auditable pair fingerprints to evaluate how models predict protein-protein binding.
Researchers have developed mmFEA, a computational framework that estimates relative metabolic reaction rates by combining gene expression or protein abundance with metabolite measurements through Michaelis-Menten enzyme kinetics.
Researchers have built RipplePLM, a computational model that generates natural language descriptions of how single-amino-acid changes disrupt protein behavior.
An analysis of approximately 2 million PubChem bioassays found that more than 99.9% lack standardized ontology tags for assay formats and detection technologies, creating major hurdles for training molecular property models.
Researchers designed an algebraic counting field model to detect cryptic binding pockets, which are hidden drug-binding sites that only become visible when a ligand attaches to a protein.
A systematic benchmark evaluated 4-bit quantization and low-rank adapter fine-tuning, known as QLoRA, across seven protein language models including ESM-2, ProtT5, and Profluent-E1.
CPathOGen uses conditional latent diffusion to generate paired synthetic histology images with independent controls for cell layout, nuclear shape, and tissue stain colors.
Fold'EM combines generative structural priors directly with raw cryo-electron microscopy particle images to infer atomic models, skipping the usual steps of building a 3D density map and fitting a model into it.
Computational biologists developed SCITRAM, a desktop tool that infers gene regulatory networks from single-cell RNA sequencing and simulates transcription factor activation cascades to identify master control proteins.
Researchers ran AlphaFold2 and AlphaFold-Multimer across 2,812 viral proteomes from 23 viral families, generating high-confidence models for 5,279 heterodimers and 2,749 homodimers.
An automated research platform named iCloudBiofoundry links AI agents that plan scientific protocols from literature with robotic laboratory hardware that executes experiments autonomously.
Proton-PottsMPNN treats protonated and deprotonated amino acids as distinct sequence tokens, allowing researchers to design proteins that bind their targets only at specified acidic pH levels.
A transformer model trained on 8.8 million experimental yeast-display binding measurements predicts whether the SARS-CoV-2 spike receptor-binding domain can engage ACE2 receptors from diverse animal species.
DeepUMQA-Selection pairs single-model accuracy scores with shape-recognition structural alignment to pick top predicted multi-protein structures from large candidate pools.
Researchers built AREION, a transformer-based brain foundation model that learns functional magnetic resonance imaging representations regardless of which brain atlas was used to define anatomical regions.
Researchers applied ChromBPNet, a deep learning model trained on single-cell chromatin accessibility profiles, to evaluate how noncoding type 1 diabetes variants alter gene regulation in pancreatic islets.
Computational biologists developed RAMP, an unsupervised alignment framework designed to harmonize shared protein readouts across cytometry, sequencing, and spatial proteomics platforms.
Researchers built BioEvidence, an interface that feeds structural and evolutionary context from standard biology tools into frozen large language models to predict how protein mutations affect function.
Standard differential expression tests such as the Wilcoxon rank-sum test, the default marker-discovery method in toolkits like Seurat and Scanpy, systematically rank genes by cluster size rather than biological specificity.
Researchers have developed Multi-ISM, a framework that models every single-letter DNA mutation across large genomic regions using 45 times fewer computer evaluations than standard exhaustive methods.
Computational biologists paired large language model agents with quantitative systems pharmacology to systematically eliminate flawed hypotheses before designing clinical trials.
Researchers introduced RetroGEF, a flow-based generative machine learning model that predicts single-step chemical retrosynthesis routes to produce target small molecules.
Researchers built CellMSA, a single-cell foundation model that borrows the concept of multiple sequence alignments in proteins to capture gene relationships across different cell types and batches.
Researchers built Mendel, an AI foundation model trained from scratch on human whole-genome sequences using a training objective focused on differences between individual genomes rather than invariant sequence.
CycloCross uses a cycle-consistent adversarial network (CycleGAN) architecture to translate single-cell RNA sequencing data between species without requiring strict one-to-one gene matches.
Researchers built CHACAM, a supervised attention model that combines single-cell gene expression with known ligand-receptor pairs and physical cell contact maps to annotate early roundworm embryo cells.