A deep auto-encoder model for gene expression prediction.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene expression links genotypes to traits and is influenced by genetic variants.
• Understanding genetic contributions to gene expression is crucial in genomics.

Purpose of the Study:

• To develop a deep learning model for predicting gene expression from SNP genotypes.
• To assess the contribution of genetic variants to gene expression changes.

Main Methods:

• A novel deep auto-encoder model, MultiLayer Perceptron and Stacked Denoising Auto-encoder (MLP-SAE), was developed.
• The model utilizes stacked denoising auto-encoder for feature selection and multilayer perceptron for backpropagation.
• Dropout was incorporated to prevent overfitting and enhance model performance.

Main Results:

• The MLP-SAE model with dropout outperformed Lasso, Random Forests, and MLP-SAE without dropout on yeast genomic datasets.
• Gene expression levels predicted solely from genotypes using the MLP-SAE model aligned well with observed expression patterns.

Conclusions:

• A deep auto-encoder model effectively predicts gene expression from SNP genotypes.
• Deep learning is suitable for building predictive models in genomics, particularly for understanding genotype-phenotype relationships.
• Deep learning models are expected to play an increasingly significant role in genomic data interpretation.