Computational genomics / regulatory genomics · BEGINNER LESSON
DeepSEA
DeepSEA is a deep learning model that predicts several functional genomic features, such as transcription factor binding, DNase I hypersensitivity, and histone modifications, directly from a DNA sequence. It was an early proof that a convolutional neural network can learn sequence patterns linked to chromatin features without hand-designed rules.
What DeepSEA is
DeepSEA is a pioneering convolutional neural network (CNN), which is a layered machine-learning model that scans data for local patterns. In this case, the input is a DNA sequence, and the measured outputs are functional genomic annotations: transcription factor binding, DNase I hypersensitivity, and histone modifications. DNase I hypersensitivity means a DNA region is easier for the DNase I enzyme to cut, which often marks open, active chromatin. Histone modifications are chemical changes on histone proteins around which DNA is wrapped.
[1] Personalized gene expression prediction in the era of deep learning [2] Deep learning for regulatory genomics: a survey of models ... [3] Table 1.
How it works
DeepSEA was trained on 1,000 base-pair DNA inputs from reference-genome data. It learns sequence patterns associated with chromatin features, so it can assign predictions from raw DNA rather than requiring preset biomarkers. This made it an early example of the sequence-to-function prediction paradigm in regulatory genomics.
[1] Personalized gene expression prediction in the era of deep learning [2] Deep learning for regulatory genomics: a survey of models ...
What it showed and why it mattered
DeepSEA demonstrated that deep learning could predict multiple chromatin-related outputs from sequence alone. That result helped launch later models that tried to improve output detail, genomic context length, and resolution for variant interpretation. In the cited summary table, DeepSEA is described as a simple CNN with strong performance on large datasets, but also with black-box interpretability limits.
[1] Personalized gene expression prediction in the era of deep learning [2] Deep learning for regulatory genomics: a survey of models ... [3] Table 1.
Limitations
DeepSEA was an important proof of concept, but its original design output one value per chromatin feature for the whole 1 kb region, so it did not give explicit position-by-position detail inside the sequence. The broader literature also notes that models trained mainly on reference or endogenous sequences may not capture individual genetic variation well, and overfitting and limited data diversity remain concerns in this field.
[1] Personalized gene expression prediction in the era of deep learning [2] Deep learning for regulatory genomics: a survey of models ...
SEE THE IDEA
How DeepSEA maps DNA sequence to chromatin-feature predictions
A 1,000 base-pair input segment is provided to the model.
The CNN looks for local sequence patterns linked to regulatory activity.
Outputs include transcription factor binding, DNase I hypersensitivity, and histone modifications.
The original model gives coarse whole-region output and limited spatial detail.
sequence input -> pattern learning -> functional genomic predictions -> interpretation limits
DeepSEA learns from raw DNA to predict chromatin-related signals, but the original version has coarse resolution.
TRY IT YOURSELF
Test your understanding
Choose an answer, then check the explanation.
PUT IT TOGETHER
Match each concept to its meaning
Drag a label onto its description, or select a label and then select a description. This also works with a keyboard.
4 pairs to match
Sources and further reading
This general lesson was generated from public source material. It explains a concept; it does not interpret a private paper or provide medical advice.