BiRGRN

BiRGRN infers gene regulatory networks from time-series single-cell RNA sequencing (scRNA-seq) data by detecting causal regulatory relationships among genes.

Key Features:

• Bidirectional recurrent neural network (Bi-RNN): Uses a Bi-RNN to capture complex, non-linear, and dynamic interactions among genes.
• Neuron-to-gene mapping: Maps neurons to genes and interprets connections between layers as candidate regulatory relationships.
• Regression formulation: Transforms GRN inference into a regression problem that predicts gene expression at subsequent time points from previous time points.
• Bidirectional structure integration: Integrates results from forward and reverse inference directions to improve robustness of inferred relationships.
• Use of prior biological knowledge: Incorporates an incomplete set of prior knowledge to filter low-confidence candidate inferences.
• Evaluation on multiple datasets: Evaluated on four simulated datasets and three real scRNA-seq datasets.
• Comparative benchmarking: Compared with other state-of-the-art techniques, reporting superior performance in benchmark analyses.
• Implementation: Implemented in Python using the TensorFlow machine-learning library.

Scientific Applications:

• GRN inference from time-series scRNA-seq: Inferring dynamic gene regulatory networks from time-resolved single-cell expression data.
• Exploration of regulatory mechanisms: Investigating gene regulatory mechanisms to advance understanding of cellular processes.
• Method benchmarking: Serving as a benchmark in comparative analyses of GRN inference methods on simulated and real datasets.

Methodology:

Train a bidirectional recurrent neural network mapping neurons to genes to predict next-time-point gene expression from previous time points, integrate forward and reverse inference directions, filter low-confidence candidate edges using an incomplete set of prior biological knowledge, and implement the approach in Python with TensorFlow; evaluated on four simulated and three real scRNA-seq datasets.