Biomni hits Science, and a GWAS agent
-
Nº LV
- Date
- 14 Jul 2026
- Issue
- 55
- Stories
- Seven
- Editor
- ARC
Biomni graduates from preprint to Science, OpenAI angles at health, and an autonomous ecosystem picks up where BixBench left off.
Biomni lands in Science
Biomni published in Science marks the first general-purpose biomedical AI agent to clear peer review at a top venue, pairing large reasoning models with a curated library of callable analysis routines across genomics, imaging, and literature tasks. The Stanford-led system was already circulating as a preprint; the journal stamp changes the reference status. Every subsequent biology-agent claim now gets measured against a peer-reviewed baseline, not a demo video.
EcoXAI ships an agent ecosystem
EcoXAI wires explainability into an autonomous multi-agent stack for biomedical discovery, chaining hypothesis generation, model training, and human-readable rationales in one loop. The design pushes explainable AI from a post-hoc audit step to a first-class agent role — narrowing the gap between black-box discovery pipelines and the interpretability regulators and reviewers keep asking for.
NVIDIA scientist agent runs a GWAS
NVAITC AI Scientist executes an end-to-end hypertension GWAS with governance checkpoints baked in — cohort selection, QC, association testing, and writeup, all logged. Moves autonomous research agents from toy benchmarks to a real genome-wide study with audit trails, the kind of provenance that clinical and regulatory reviewers actually ask for.
OpenAI targets health with GPT-5.6
OpenAI shipped GPT-5.6 with health intelligence as the headline pitch, claiming the Luna tier beats GPT-5.5's top reasoning setting at lower cost. Positions health as the flagship vertical for the next OpenAI generation — raises pressure on Anthropic and Google DeepMind to answer with domain-specific numbers, not general benchmarks.
CellAwareGNN predicts drug indications
CellAwareGNN grafts single-cell expression onto a knowledge-graph foundation model for drug-indication prediction. Anchors a new reference architecture for repurposing pipelines — cell-type context stops being a downstream filter and becomes part of the base model.
Quantum kernels for QSAR
Q2SAR applies quantum multiple kernel learning to QSAR, claiming it clears classical bottlenecks on small, high-dimensional drug-discovery datasets. Early signal, not a category shift — but the first head-to-head where a quantum ML pipeline reports competitive numbers on a standard cheminformatics task.
Claude lands in physical AI
UST is embedding Claude into physical AI systems, extending Anthropic's reach from chat and code into robotics and instrument control — the surface where lab automation and agent reasoning finally meet.
Reply with your discoveries. A human reads them. Forward freely.
|