Deep Learning Methods for Predicting Disease Status Using Genomic Data.

Area of Science:

• Genomics
• Machine Learning
• Personalized Medicine

Background:

• Predicting complex human disease status from genomic data is crucial for personalized medicine.
• The curse of dimensionality often hinders the performance of machine learning algorithms with high-dimensional genomic datasets.
• Deep learning offers a promising approach for feature extraction from complex, high-dimensional data.

Purpose of the Study:

• To review the application of deep learning in predicting disease status from genomic data.
• To evaluate the effectiveness of deep learning methods in overcoming the curse of dimensionality in genomic analysis.
• To compare deep learning approaches with existing prediction methods for disease status.

Main Methods:

• Literature search for relevant articles on deep learning and genomic data for disease prediction.
• Analysis of studies employing auto-encoders for dimensionality reduction of genomic data.
• Application of state-of-the-art machine learning algorithms on low-dimensional genomic representations.

Main Results:

• Four relevant articles were reviewed, all utilizing auto-encoders for genomic data dimensionality reduction.
• Deep learning approaches demonstrated superior performance in disease status prediction compared to transcript-wise screening and principal component analysis.
• The reviewed studies highlight the potential of deep learning in enhancing the accuracy of genomic data-driven disease prediction.

Conclusions:

• Deep learning, particularly using auto-encoders, shows significant promise for predicting disease status from high-dimensional genomic data.
• These methods effectively address the curse of dimensionality, outperforming conventional techniques.
• Further research into deep learning's limitations and improvements is warranted for advancing personalized medicine.