scDLC

scDLC classifies large-scale single-cell RNA sequencing (scRNA-seq) data using long short-term memory (LSTM) recurrent neural networks to identify cell types and patterns in gene expression.

Key Features:

• Deep Learning Framework: Core built on long short-term memory (LSTM) recurrent neural networks that capture long-term dependencies and complex relationships among genes in scRNA-seq datasets.
• Scale Invariance: Operates as a scale-invariant method and does not require preprocessing steps for data scaling or normalization.
• Independence from Data Distribution Assumptions: Does not rely on specific distributional assumptions such as Poisson, negative binomial, or zero-inflated Poisson used by PLDA, NBLDA, and ZIPLDA.
• Focus on Feature Gene Dependencies: Emphasizes modeling dependencies among the most significant feature genes within the LSTM architecture.
• Performance Evaluation: Shown by simulation studies and analyses of real-world datasets to consistently outperform existing methods across large sample settings.

Scientific Applications:

• Disease Diagnosis and Medical Research: Classifies cells in scRNA-seq datasets to support disease diagnosis and related medical research.
• Cellular Heterogeneity Analysis: Resolves and characterizes cellular heterogeneity by classifying cells based on gene expression profiles.
• High-Throughput scRNA-seq Studies: Applies to large-scale, high-throughput scRNA-seq datasets for cell-type identification and pattern discovery.

Methodology:

Training long short-term memory (LSTM) recurrent neural networks on scRNA-seq gene expression matrices without prior distributional assumptions or normalization, modeling dependencies among selected feature genes.