Sequential regulatory activity prediction across chromosomes with convolutional neural networks.

Area of Science:

• Genomics
• Computational Biology
• Machine Learning

Background:

• Predicting phenotypic outcomes from genotypes is crucial for understanding genomic function and improving human health.
• Accurate prediction of gene expression from DNA sequence is a key challenge in genomics.

Purpose of the Study:

• To develop a machine-learning system for predicting cell-type-specific epigenetic and transcriptional profiles from DNA sequence alone.
• To identify regulatory elements and predict gene expression using convolutional neural networks.

Main Methods:

• Utilized convolutional neural networks (CNNs) to analyze DNA sequences.
• Trained the model to identify promoters and distal regulatory elements.
• Synthesized information from regulatory elements to predict gene expression levels.

Main Results:

• The machine-learning system successfully predicted cell-type-specific epigenetic and transcriptional profiles.
• Model predictions for the impact of genomic variants on gene expression correlated well with causal variants in human expression quantitative trait loci (eQTLs).

Conclusions:

• The developed system effectively predicts gene expression and regulatory element function from DNA sequence.
• This approach can generate mechanistic hypotheses for fine-mapping disease loci and understanding the functional impact of genomic variants.