An Overview of Deep Generative Models in Functional and Evolutionary Genomics

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Deep learning is widely adopted in genomics.
• Deep generative models (DGMs) are emerging as a powerful methodology in the broader field of genomics.
• DGMs can learn complex genomic data structures and generate novel, realistic genomic instances.

Purpose of the Study:

• To review the principles of generative modeling and prevailing DGM architectures.
• To present conceptual applications and notable examples of DGMs in functional and evolutionary genomics.
• To discuss potential challenges and future research directions for DGMs in genomics.

Main Methods:

• Introduction to generative modeling concepts.
• Overview of two prevalent deep generative model architectures.
• Conceptualization of DGM applications in genomics.

Main Results:

• DGMs facilitate the generation of novel genomic data that mirrors original dataset characteristics.
• DGMs enable dimensionality reduction by mapping genomic data to a latent space.
• DGMs can be utilized for prediction tasks through learned mappings or supervised/semi-supervised designs.

Conclusions:

• Deep generative models are versatile tools for genomic data analysis and generation.
• DGMs hold significant potential for advancing functional and evolutionary genomics research.
• Addressing challenges and exploring future directions will further unlock the capabilities of DGMs in genomics.