Home/Research and explanations

Research and explanations

Virtual cells, protein design, automated laboratories and the experiments behind cell repair.

15 articles

Protein pocket holding a copper molecular candidate, with other candidate shapes nearby.

Explainer4 min read

What AI does in drug discovery

AI can help researchers choose a target, search chemical space and design molecules. Those are different tasks, with different tests of success.

Read article

Cutaway cell with an intact outer membrane and organelles surrounding copper forms inside its nucleus.

Explainer4 min read

How partial cellular reprogramming works

Reprogramming factors change how cells use their genes. Researchers are testing whether a controlled period of expression can restore selected functions while preserving cell identity.

Read article

Cell samples arranged beside a sculptural calibration instrument with nested plates.

Explainer4 min read

What a younger ageing-clock reading can tell us

An ageing clock converts biological measurements into a prediction. Its training target determines what that prediction means. A lower reading after an intervention still needs to be tested against lasting health outcomes.

Read article

A molecular model, an assay plate and blank observation records connected across a tabletop.

Explainer4 min read

Why faster discovery still faces slow clinical development

AI can search for molecules quickly. Clinical studies still need time to measure exposure, effects and harms. Time saved in discovery is only part of a development programme.

Read article

A lipid-like carrier containing an RNA strand beside a cell membrane and an opened internal vesicle.

Explainer4 min read

Getting rejuvenation machinery into the right cells

A delivery system must reach the intended cells and control exposure. Lipid nanoparticles, viral vectors and local administration solve different parts of that problem.

Read article

A figure walking on an assessment platform beside a tissue specimen with simplified cells.

Explainer4 min read

What would establish human rejuvenation?

A claim about human rejuvenation needs a defined change, a credible comparison and enough observation to establish benefit and harm. Tissue markers, functional outcomes and longer follow-up answer different questions.

Read article

Lung cutaway beside an abstract molecular candidate approaching a protein-shaped target.

Investigation6 min read

Rentosertib: discovery, trials and ageing clocks

Rentosertib is being tested for idiopathic pulmonary fibrosis after AI-assisted drug discovery. An exploratory ageing-clock analysis examines changes in participants’ blood proteins.

Read article

Illustrated eye cutaway with simplified retinal layers and an optic nerve leaving the back of the eye.

Investigation6 min read

ER-100 reaches human testing: what the trial can establish

ER-100 tests controlled expression of reprogramming factors in an early human study of optic neuropathies. The study assesses safety and vision; an interim presentation is scheduled for 8 October 2026.

Read article

Illustrated knee cutaway with cartilage surfaces and an inset containing several simplified cells.

Investigation6 min read

UBX0101: a senolytic hypothesis meets a negative trial

UBX0101 failed its prespecified pain endpoint in a 183-participant knee osteoarthritis trial. Development stopped. The result does not establish why the treatment failed.

Read article

Five illustrated cells connected to a panel containing simplified computational representations.

Explainer5 min read

What is a virtual cell, and what can it predict?

A virtual-cell model predicts a measurement after researchers change a cell. Its usefulness depends on which experiment it can predict, which examples it has already seen and whether the prediction survives a new biological context.

Read article

A cell population beside three different computational representations.

Investigation5 min read

How should we test an AI model of a cell?

A 2025 benchmark found that tested deep-learning models failed to beat simple predictions. An October 2026 study shows that scoring choices can conceal useful performance. Their comparison turns on the task, the controls and what a metric rewards.

Read article

Grouping a folded protein, a regulatory DNA segment, a sequence strip and a population of cells.

Explainer5 min read

AlphaFold, AlphaGenome, Evo 2 and State: what each model does

AlphaFold predicts molecular structures, AlphaGenome predicts regulatory measurements from DNA, Evo 2 models genomic sequences and State predicts cellular responses. Each needs a different experimental test.

Read article

A folded protein binder approaching a receptor with a matching surface.

Investigation5 min read

When an AI-designed protein works in the laboratory

BindCraft and RFdiffusion3 have produced proteins that pass physical tests. Interpreting each result requires the number tested, the assay and the biological task. Binding, catalysis and targeted delivery require different evidence.

Read article

A pipetting head above an assay plate, with a looping path around the apparatus.

Investigation5 min read

What happens when AI chooses the next laboratory experiment?

An automated laboratory becomes a learning system when measured results determine the next experiments. Cell-free protein synthesis and enzyme engineering show what these systems have achieved, how people contributed and which comparisons remain missing.

Read article

An islet-like cluster of cells surrounded by several immune cells.

Investigation5 min read

Replacing islets: the cells work, but how do they survive?

Zimislecel and UP421 test cell replacement in type 1 diabetes. Zimislecel has produced insulin independence in a small study with immunosuppression. UP421 tests whether edited donor cells can survive without it.

Read article