AGImpute

AGImpute imputes dropout events in single-cell RNA sequencing (scRNA-seq) data to recover missing gene expression values and improve downstream analyses such as clustering, marker gene identification, and trajectory inference.

Key Features:

• Dynamic Threshold Estimation: Employs a dynamic threshold estimation strategy to differentially estimate the number of dropout events across cells, accounting for variability from sequencing protocols, cell types, and batch effects.
• Hybrid Deep Learning Model: Integrates an Autoencoder with a Generative Adversarial Network (GAN) to impute identified dropout events by capturing latent representations and generating realistic data distributions.
• Comprehensive Validation: Validated against seven state-of-the-art dropout imputation methods using two simulated datasets and seven real scRNA-seq datasets, and shown to impute fewer dropout events compared to other methods.
• Enhanced Downstream Analysis: Improves clustering performance, cell-specific marker gene identification, and trajectory inference, including time-course datasets.

Scientific Applications:

• Cell Type Identification: Improved clustering performance aids in distinguishing between different cell types.
• Marker Gene Discovery: Accurate imputation facilitates identification of genes specific to particular cell populations.
• Developmental Biology and Trajectory Inference: Enhanced trajectory inference supports analysis of cellular development and differentiation, including time-course studies.

Methodology:

AGImpute first estimates dropout events per cell using a dynamic threshold that considers sequencing protocol, cell type, and batch variability, and then imputes the identified dropouts using a hybrid model combining an Autoencoder and a Generative Adversarial Network (GAN).