Techbio Today

The front page of AI-driven biology

TechBio Today, today’s edition

Today’s spotlight

Selected across News, Publications, and Preprints

  • News

    AstraZeneca taps insitro's AI platform for liver and heart drug targets

    AstraZeneca is paying insitro $40 million upfront, with up to $1.46 billion in potential milestones that bring the deal's headline value to $1.5 billion, to identify drug targets for metabolic liver disease and heart failure. AstraZeneca secures exclusive worldwide rights to advance any successful programs emerging from the alliance, which follows previous discovery partnerships insitro formed with Bristol Myers Squibb, Gilead, and Eli Lilly.

    Fierce Biotech, also reported by ReutersEndpoints News (subscription required)
  • News

    Machine learning identifies brain-spliced microexons linked to autism

    Researchers from the University of Toronto, the Sloan Kettering Institute, King's College London, and the Francis Crick Institute applied a computational tool named MicroExonator to over 100 terabytes of RNA sequencing data, uncovering more than 1,500 tiny protein-coding segments known as microexons. A trained machine learning model predicted which rare genetic variants disrupt the regulatory code that controls microexon inclusion in brain transcripts. The team found that predicted splicing-altering variants clustered heavily among autistic individuals who previously lacked an identified genetic contributor to their condition.

    GEN
  • News

    Illumina bundles multiomics software and splice-predicting AI into new BioInsight platform

    Illumina announced the launch of BioInsight, a unified software suite that combines its multiomics analytics, artificial intelligence tools, and biological datasets. The software includes SpliceA12, an updated genomic AI model that identified 17% more disease-linked splicing variants than competing tools in tests across Genomics England patient cohorts. Splicing alterations misdirect how cells piece together RNA transcripts, making them a major driver of rare inherited disorders, familial cancers, and drug response differences.

    GEN
  • Publications

    Upgraded diffusion model scaffolds intricate enzyme active sites from scratch

    University of Washington researchers have introduced RFdiffusion3, a generative model that uses free-atom diffusion to build custom protein and RNA structures around precise geometric constraints. Tested across functional enzyme challenges, the system designed synthetic phosphotriesterases that coordinate two metal ions and zinc-dependent proteases tailored to cut transthyretin, a protein implicated in amyloid diseases. Both enzyme campaigns yielded functional catalysts without iterative rounds of laboratory optimization, producing active hits within 96 tested candidates.

    Science
  • Preprints

    Synthetic negative examples scramble rankings of protein interaction models

    Because laboratory tests rarely document proteins that fail to interact, computational biologists evaluate interaction models against randomly paired non-interacting proteins, an approach that systematically warps performance metrics. Testing four sequence-based algorithms across six different ways of constructing negative examples showed that altering only the test negatives changed the area under the precision-recall curve by up to 0.26, reshuffling which architecture ranked first. When negative pairs were closely matched in representation space using protein language models, all language model architectures dropped to near-chance accuracy, and carefully constructed training negatives underperformed uniform random sampling in 108 of 120 external comparisons.

    bioRxiv

This week’s top AI x bio headlines

From News, the past seven days

  1. Iambic Therapeutics raises $150 million in Nasdaq debut

    BioPharma Dive

  2. Colossal Biosciences unveils AI platform to guide gene edits for dodo de-extinction

    TechCrunch

  3. Zealand Pharma and Generate partner on AI-designed obesity drugs in $1.2 billion pact

    Reuters

  4. DeepMind watermarks AI-designed proteins without reducing target binding

    Yahoo Finance

  5. GSK licenses Chai Discovery's protein-design models after lab testing

    Fierce Biotech

View all news →

Publications

View all →
  • Language models pull standardized rare-disease symptoms directly from patient notes

    Researchers built RARE-PHENIX, an automated framework that reads unstructured clinical text, standardizes symptoms into formal Human Phenotype Ontology terms, and ranks them by diagnostic importance.

    npj Digital Medicine
  • AI scans link brown fat activity to lower psychiatric disease rates and flag candidate drugs

    An artificial intelligence system quantified human brown adipose tissue activity across clinical scans, revealing strong associations between active brown fat, improved metabolic profiles, and reduced rates of psychiatric illness.

    Nature
  • Graph-based method parses tissue architecture across spatial scales

    SpatialZoomer uses spectral graph theory to analyze spatial transcriptomics data simultaneously at individual cell resolution and across broad tissue domains.

    Nature Biomedical Engineering
  • Billion-spectrum proteomics dataset powers fast model for reading peptide sequences

    Researchers have released MassNet, an AI-ready training dataset containing 1.5 billion tandem mass spectrometry spectra across 35 species, stored in a tensor format designed for deep learning.

    Nature Methods

View all publications →

Preprints

View all →

View all preprints →

TechBio Today’s AI x Bio Feed

Follow @TechBioTodaySee the list →

Upcoming conferences, seminars, and events

Send feedback