The front page of AI-driven biology
Thursday, October 15, 2026Thu, Oct 15
Updated 7:01 PM PT
Researchers built EpiZoo, a 2.6-billion-parameter mixture-of-experts model trained on 20.9 million single cells from multiple species to analyze chromatin accessibility, the physical opening of DNA for gene control.
AbbVie, argenx, Lundbeck, and Takeda have joined Ginkgo Bioworks and Apheris to assemble a standardized dataset of 10,000 antibodies aimed at training predictive AI models for drug manufacturing and formulation.
Researchers have built CROWN, a foundation model designed to analyze cell samples across diverse organs, diseases, and clinical settings.
Researchers built NexuST, a biological foundation model that interleaves gene-level patterns with tissue-level spatial context during pretraining.
Researchers built RNASeek, a 1.6-billion-parameter foundation model trained on transcriptomes across multiple phyla to generate and optimize regulatory RNA sequences.
Fine-tuning the open-source co-folding model Boltz-1 on ten proprietary X-ray structures of Werner helicase enabled it to predict both an unseen allosteric binding pocket and the large inactive conformational shift induced by small molecules.
A benchmark evaluating ten general-purpose single-cell foundation models, three aging-specific models, and classical baselines across 2.5 million cells found that model utility depends heavily on the specific biological task.
CheckAMG combines standard functional annotation with a fine-tuned genomic language model to find auxiliary viral genes, which phages and other viruses use to rewire host metabolism, physiology, and regulation.
General protein language models such as ESM-2 and ProtGPT2 fail to separate true T-cell receptor binders from synthetic negative controls, according to a benchmark evaluation.
A team developed Cell-Lens, a training-free framework that structures single-cell data into typed blocks containing paired measurements and marker programs rather than simple ranked gene lists.
CAMP-14-3-3 is a computational framework that predicts binding sites on 14-3-3 proteins, which regulate cellular signaling by docking to phosphorylated serine and threonine residues.
A team led by Tsinghua University introduced MicroGlot, a genomic language model pretrained on 378.3 billion nucleotides across 99,700 microbial species.
Researchers have developed UpTCR, a progressive transfer-learning model designed to predict whether T-cell receptors will bind specific peptide targets presented by human leukocyte antigen molecules.
Researchers introduced Minerva, an alignment-free framework that uses a genome language model to predict RNA base-pairing, protein contacts, and repeat sequences as two-dimensional interaction maps.
Researchers guided large language models to suggest specific chemical modifications that make computationally generated molecules easier to synthesize in the laboratory.
Researchers have built a molecular foundation model that predicts cryptic binding pockets directly from unliganded, static protein coordinates without running structural simulations.
Anthropic has launched its newest frontier model, Opus 5.5, which cuts output token pricing to $20 per million tokens compared to $25 for its predecessor.
A language-model-guided computational tool identified specific guanine motifs that govern where RNA polymerase II pauses near gene promoters.
Researchers introduced PREpiBind, a dual-stream neural network that tests ten different protein representations for predicting how peptides bind class II major histocompatibility complex (MHC-II) proteins, which present antigens to immune cells.
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.
Scientists used the protein language model Lucaprot and AlphaFold2 structure modeling to detect two previously unknown RNA viruses in RNA sequencing data from a 14,300-year-old permafrost-preserved Pleistocene wolf.
Researchers built a language model named DeepAir-Cov19 to spot SARS-CoV-2-targeting T-cell receptors, achieving an area under the curve of 0.94.
Researchers built CELL-FM, a bidirectional generative AI framework that predicts fluorescence microscopy images directly from protein sequences and cellular context, and can also work in reverse.
Anthropic has set up a physical biology laboratory to run wet-lab experiments that validate predictions made by its artificial intelligence models, life sciences head Eric Kauderer-Abrams told Reuters.
Researchers built OpenLipid, a workflow that uses large language models to identify target lipids from data-independent acquisition mass spectrometry.
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.
An update to the open-source ModelSEED Biochemistry Database expands the resource to approximately 46,000 compounds and 56,000 mass- and charge-balanced reactions for whole-cell metabolic modeling.
Novo Nordisk has partnered with AI startup Anthropic to deploy its Claude model across early drug discovery and research operations.
Novo Nordisk plans to deploy Anthropic's Claude Science model to assist with scientific reasoning across drug research and development and agentic software tasks.
A computational pipeline called Rider combines protein language models with structural validation to identify divergent RNA viruses from more than 10,000 environmental metatranscriptomes.