DeepMicroGen: a generative adversarial network-based method for longitudinal microbiome data imputation.

Area of Science:

• Microbiome Research
• Computational Biology
• Bioinformatics

Background:

• The human microbiome significantly impacts health and is linked to various diseases.
• Longitudinal microbiome studies are crucial for understanding disease associations but suffer from missing data due to varying sample sizes and time points.
• Deep generative models, particularly Generative Adversarial Networks (GANs), show promise for data augmentation and imputation in time-series data.

Purpose of the Study:

• To propose DeepMicroGen, a novel GAN-based model for imputing missing microbiome samples in longitudinal studies.
• To evaluate DeepMicroGen's performance against traditional imputation methods.
• To assess the impact of DeepMicroGen imputation on clinical outcome prediction, specifically for allergies.

Main Methods:

• Developed DeepMicroGen, a bidirectional recurrent neural network-based GAN.
• Trained the model on the temporal relationships within microbiome observation data.
• Compared DeepMicroGen with standard baseline imputation methods using simulated and real-world datasets.

Main Results:

• DeepMicroGen demonstrated superior performance compared to standard imputation methods, achieving the lowest mean absolute error.
• The model successfully imputed missing values in both simulated and real microbiome datasets.
• Utilizing DeepMicroGen for imputation improved the accuracy of a clinical outcome classifier for allergies trained on incomplete longitudinal data.

Conclusions:

• DeepMicroGen offers an effective solution for handling missing data in longitudinal microbiome studies.
• The proposed model enhances the quality of microbiome data analysis, leading to improved clinical outcome predictions.
• DeepMicroGen is publicly available, facilitating its adoption in microbiome research.