Assembling the brief…
The front page of AI-driven biology.
Social
A week of AI × biology findings, tools and discussion, selected for substance and a range of voices. Newest posts first.
Turner NextGen AI argues that sequence-to-outcome models can miss the biological relationships that make a mutation matter. The post proposes richer representations of structure and regulation as a route beyond correlation, a conceptual critique rather than a new experimental result.
View post on XEric Topol highlights early autoimmune-neurology results from in vivo B-cell depletion and contrasts the approach with labor-intensive ex vivo methods. The post points to the practical promise of engineering therapeutic cells inside the body, but it does not establish durability or safety at scale.
View post on XDerya Unutmaz highlights Insilico Medicine's release of five specialized AI models for drug discovery. They add tools for researchers to evaluate, but enthusiasm for the launch should not be confused with demonstrated progress in patients.
View post on XMohammed AlQuraishi points readers to David Pearlman's assessment of OpenBind-0, the open molecular-structure model. The post is a reading recommendation for researchers evaluating structure-based AI, not a new model launch or an independent validation of its performance.
View post on XKevin Yang announces the Cell Systems publication of his team's biologically inspired models for predicting molecular smell. The work connects computational modeling with sensory biology, an AI × Bio application beyond the usual protein-structure and drug-discovery tasks.
View post on XPardis Sabeti highlights BroadBox, an effort to test AI agents through controlled scientific experiments. The linked preprint finds that agents can improve scores in biological tasks without understanding the underlying rules. That distinction is useful when evaluating AI scientists.
View post on XBiohub reports that its AI-designed fluorescent proteins, called Rhobins, function in living cells, with results published in Cell. The work shows how generative protein design could expand the molecular tools researchers use to observe biology.
View post on XAlex Zhavoronkov highlights PathStAR, which uses pathology foundation-model features to map age-related changes across human tissues. His post argues that tissue architecture can complement molecular aging clocks; the linked Nature Aging analysis is observational and does not demonstrate an age-reversing treatment.
View post on XNimi Vashi argues that enthusiasm for omics and high-throughput data has diverted attention from simpler, fundamental biological questions. For AI × Bio researchers, the provocation is whether collecting more data is helping answer the right questions; this is a viewpoint, not an empirical verdict on omics.
View post on XAlex Zhavoronkov highlights RadMatch, a research method that evaluates AI-written radiology reports by matching individual findings and identifying specific errors. The linked preprint aims to make evaluation more auditable than a single score; it does not establish that an AI system is safe for clinical deployment.
View post on XThe Broad Institute shares a method that engineers cells to release RNA-containing particles, allowing researchers to sample gene activity over time. Such longitudinal measurements could support models of changing cell states, but the post describes a research platform, not a clinically validated test.
View post on XWill Hua reports that OpenDDE 1.1.1 adds Apple Silicon support, lowers inference memory use and improves reproducibility. The changes could make molecular-modelling workflows easier to run on a wider range of machines; these are developer-reported engineering improvements, not evidence of better drug candidates.
View post on XJustin Blethrow says Plasmidsaurus now includes mycoplasma checks in RNA-seq orders, automatically for academic customers and by opt-in for commercial users. The added screen could help flag compromised experimental data, but he cautions that it does not replace conventional contamination testing.
View post on XAh-Ram Kim challenges how Anthropic communicates and applies biological-safety restrictions, using a basic question about proteins as an example. The post raises a workflow question for biologists using AI: how to keep routine research useful while setting proportionate safety boundaries; it is criticism, not a systematic test of the model.
View post on X