Skip to content

What new research actually found.

We explain what new science and AI papers found and where the work goes next.

Read the latest breakdownGet the daily digest

Free daily email. Unsubscribe anytime.

Latest breakdowns

Browse all breakdowns
Top story
AI methods

Evo 2 Generates Candidate Phage Genomes From Examples Without Fine-Tuning

Genome language models usually need extra training before they can design a new kind of sequence. A preprint reports that Evo 2 can be steered by examples placed in its prompt.

Read the breakdown
Illustration of small round T cells next to two larger lymphoma cells, with three short receptor links joining one T cell to a tumor cell.
Cell therapy

Lymphoma Study Links Lasting CAR T Responses to Host CD8 T Cells

A preprint on large B-cell lymphoma suggests that engineered T cells start the attack and the patient's own T cells help sustain it.

Read the breakdown
Illustration of a branched sugar chain drawn as squares, circles and diamonds, attached at one point to the surface of a folded protein.
AI methods

GlycoWeave Predicts Bound Sugar Chains with Chemical Constraints

Sugar chains on protein surfaces keep defeating general structure predictors. A new preprint searches for the bound pose without ever altering the sugar's chemistry.

Read the breakdown
Teal, gold and gray crystal lattices follow a curved timeline beside archival sketches above a pale mountainous landscape.
AI methods

Learning Materials' Reporting History Improves Synthesizability Rankings

A model that predicts which inorganic compositions can be made does better when it also learns the order in which known materials were found.

Read the breakdown
Six connected checkpoints along a winding path, each accompanied by a medical record and an incomplete geometric puzzle.
Clinical AI

Language Models Struggle with Staging in a Nasopharyngeal Cancer Benchmark

A model’s answer at diagnosis may say little about its reliability at the next stage of cancer care.

Read the breakdown

How a breakdown
works

A study of DNA editing tags shows how a useful lab shortcut can change the result.

Read the complete cPRIME breakdown

Human cell lines

A DNA tag can change how RNA is processed

The cPRIME screen used marker tags to distinguish edits from sequencing errors. Some edit-and-marker pairings also created splice signals, changing the RNA result they were meant to help measure.

Variants measurable above background

With markersnearly 90%
Without markersabout 10%

Human cell lines

Markers made more variants measurable

In leukemia cell cultures, nearly 90% of variants were quantifiable above background with markers, versus about 10% without. Removing a marker reduced the predicted splice effect, though one variant still lowered RNA reads.

Variants measurable above background

With markersnearly 90%
Without markersabout 10%

Human cell lines

Find out why some RNA forms are depleted

The authors propose long-read transcript sequencing to examine altered RNA forms. Blocking RNA decay could help distinguish transcript loss from changes in splicing.

Variants measurable above background

With markersnearly 90%
Without markersabout 10%