Research Daily Summary: What Happened on October 11, 2026

Missed NHS checks linked to higher risks of death and cardiovascular events

Compared with attendees, people who missed an NHS Health Check had a 31% higher risk of death from any cause and a 9% higher risk of a first nonfatal heart attack or stroke.1 Imperial College London researchers analyzed anonymized NHS records for 925,072 eligible people from 2015 to 2025. Forty-eight percent had attended a check.1

England offers the checks every five years to eligible people aged 40 to 74 who do not have specified pre-existing conditions.1 In North West London, checks identified 7,806 people with high blood pressure within 30 days. A further 4,587 attendees received a type 2 diabetes diagnosis.1 Of those identified with high blood pressure, 55% had brought it to healthy levels within six months, according to the report.1

Researchers formally presented the analysis at the European Society of Cardiology Congress in Munich in August.1 They want to test ways to reach people who miss checks, including digital reminders and changes to invitation letters, locations or appointment times.1

AI-biology work examines protein tracing and biosecurity

Google DeepMind developed a method designed to mark AI-designed proteins for identification and tracing without affecting their function.2 A peer-reviewed study found that genome language models could generate functional bacteriophages. Some outcompeted naturally occurring phages and overcame E. coli resistance.2

RAND said some AI agents can alter DNA sequences to bypass current screening systems, and may help non-experts despite safeguards, including by misrepresenting their identity.2 A separate analysis disputes that AI will readily enable biological weapons development, citing the practical expertise and experimentation involved.2

A five-year Bio Action Plan roadmap proposes assessing AI-related biological risks, strengthening safeguards and coordinating work across industry, philanthropy and government.2 Research on AI biosecurity benchmarks suggests refusal rates alone are not enough to assess a model’s safety performance.2

Also worth knowing

• A

  Tamarind Bio’s Molecular Model Router uses benchmark results to help researchers choose molecular AI models, beginning with biomolecular structure prediction.2

• T

  A study trained on Pharos data through 2017 identified 5,019 proteins not yet classed as clinically validated drug targets, including 145 of 181 targets already in Phase I–III trials.3

• T

  Causal-Chemprop uses measurements from a related anchor molecule to improve predictions for molecules outside its training distribution without retraining. Results depend on anchor choice.3