Predictive design of tissue-specific mammalian enhancers that function in the mouse embryo

In plain English

The team trained DNA-sequence models, refined them with validated enhancers, and tested newly designed regulatory DNA in embryos.

What they found

• All 15 designs were active in the intended heart, limb, or central-nervous-system tissue.
• The models inferred enhancer activity from DNA sequence after transfer learning on modest validated sets.

The catch

• The experiments covered three tissues at one embryonic stage.
• The study tested individual enhancers with a standardized reporter setup.

Methods at a glance

Study type

Model development with transgenic validation

Biological system

Tissue-specific mammalian enhancers

Organism

Mouse embryos

Dataset

Mouse chromatin-accessibility data and validated human and mouse enhancers

Sample size

15 designed enhancers

Model architecture

Compact convolutional neural networks with transfer learning

Training objective

Predict and design tissue-specific enhancer activity

Code

Code availability is described by the publisher.

Data

Data availability is described by the publisher.

Tasks and metrics

Tasks: Heart, limb, and central-nervous-system enhancer activity in E11.5 embryos

Metrics: Reporter activity and tissue specificity

Baselines: Held-out genomic regions and validated enhancer sequences

Principal findings

• All 15 tested designs were active in the intended tissue.

What remains unresolved

External validation: Transgenic mouse-embryo reporter assays

• Three tissues and one developmental stage were tested.
• The assay used individual enhancers in a reporter construct.