Deep Learning Methods for Omics Data Imputation.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Missing values are a common challenge in omics data analysis, impacting data integrity.
• Traditional methods for handling missing data, such as deletion, can lead to significant data loss.
• Imputation methods offer a way to estimate and fill in missing values, preserving data for analysis.

Purpose of the Study:

• To provide a comprehensive review of deep learning-based imputation methods for omics data.
• To focus on deep generative model architectures like autoencoders, VAEs, GANs, and Transformers.
• To emphasize the application of these methods in multi-omics data imputation.

Main Methods:

• Review of existing literature on deep learning for omics imputation.
• Categorization of methods based on deep generative model architectures.
• Analysis of challenges and opportunities in the field.

Main Results:

• Deep learning models excel at capturing complex, non-linear relationships in high-dimensional omics data.
• Various deep generative architectures (Autoencoder, VAE, GAN, Transformer) are adapted for omics imputation.
• These methods show promise for handling technical variations and non-random missingness.

Conclusions:

• Deep learning offers powerful tools for addressing the complexities of omics data imputation.
• Further research is needed to overcome challenges and fully leverage deep learning's potential in multi-omics data analysis.
• The review highlights opportunities for advancing omics data imputation through deep generative models.