arboreto

Arboreto – A Gene Regulatory Network Inference Tool

Overview of Capabilities

Arboreto is a Python library specifically designed to infer gene regulatory networks (GRNs) from gene expression data. Using the GRNBoost2 and GENIE3 algorithms, it identifies regulatory relationships between transcription factors and target genes, while supporting distributed parallel computing for large-scale datasets.

Use Cases

1. Gene Regulatory Network Analysis of Single-Cell RNA-seq Data

When you need to identify cell type-specific gene regulatory networks from single-cell sequencing data, Arboreto can efficiently process expression matrices containing thousands of cells and genes, helping reveal transcriptional regulatory mechanisms under different cellular states. It supports sparse matrix formats to optimize memory usage.

2. Transcription Factor Target Gene Prediction from Bulk RNA-seq Data

For conventional transcriptomic sequencing data, Arboreto can filter a specific list of transcription factors, predict their potential target genes, and use importance scores to identify high-confidence regulatory relationships. It is suitable for comparative analyses involving multiple experimental conditions, such as control groups and different treatment time points.

3. Distributed Computing for Large-Scale Gene Networks

When the data scale exceeds the memory or computational capacity of a single machine, Arboreto supports local multi-core parallelism and remote cluster computing through the Dask framework. You can customize the number of worker processes and memory limits, or connect to an existing Dask scheduler for elastic scaling.

Core Features

1. Support for Two Algorithms (GRNBoost2 and GENIE3)

Arboreto integrates two mainstream gene regulatory network inference algorithms. GRNBoost2 is based on gradient-boosted trees and is optimized for large-scale datasets with more than 10,000 observations, making it the recommended choice for most analytical scenarios. GENIE3 uses random forests and is suitable for algorithm comparison and validation studies. Both algorithms support reproducibility through a seed parameter.

2. Flexible Data Formats and Input Support

Arboreto supports Pandas DataFrames, NumPy arrays, and SciPy sparse matrices in CSC format as inputs, accommodating different data preprocessing workflows. For array-format inputs, gene names can be specified explicitly. The output is a standard DataFrame containing three columns—transcription factor, target gene, and importance score—for convenient downstream analysis and visualization.

3. Seamless Integration with the pySCENIC Ecosystem

As an upstream dependency of pySCENIC, the regulatory networks generated by Arboreto can be directly passed to pySCENIC for cis-target motif enrichment analysis and regulon definition. It also supports converting AnnData objects, such as single-cell data processed with Scanpy, into DataFrame format, enabling a smooth transition between analysis steps.

Frequently Asked Questions

What types of data analysis is Arboreto suitable for?

Arboreto is specifically designed for gene regulatory network inference from transcriptomic data and is best suited for single-cell RNA-seq and bulk RNA-seq data. The input data should be an expression matrix with genes as columns and observations, such as cells or samples, as rows. If your data consists of other types of omics data, such as proteomics or metabolomics, or if the rows and columns of the expression matrix are transposed, Arboreto may not be the best choice.

How should I choose between GRNBoost2 and GENIE3?

For most practical applications, GRNBoost2 is recommended because it offers better speed and memory efficiency, particularly for large-scale datasets containing more than 10,000 observations. GENIE3 is suitable for comparison with and validation against existing research results, or when you need to use the classic random forest approach. The outputs of both algorithms have the same format, so they can be run in parallel and their results combined by intersection or union to improve robustness.

What computational resources does Arboreto require?

Arboreto’s resource consumption depends on the data size and algorithm selected. When run locally, it automatically uses all available CPU cores. For particularly large datasets, configuring a distributed computing environment through Dask is recommended. At least 8 GB of memory is recommended, while 32 GB or more is advisable when processing whole-genome single-cell data. If you encounter insufficient memory, you can reduce computational complexity by filtering out low-variance genes or limiting the transcription factor list.