Introduction

Beaver implements an efficient algorithm to provide more accurate transcript assembly at single-cell resolution. The datasets and scripts used to compare the performance of Beaver with other assemblers are available at beaver-test.

Installation

Clone the latest Beaver repository:

git clone https://github.com/Shao-Group/beaver.git
cd beaver

Generate configuration files:

aclocal        # Generate aclocal.m4
autoconf       # Generate configure script
autoheader     # Generate config.h.in
automake -a    # Generate Makefile.in

Configure and compile:

./configure    # Generate Makefile
make          # Compile the source code

The executable will be generated at /path/to/your/beaver/src/beaver.

Usage

./beaver -i <input-gtf-list> -o <output-prefix> [options]

A quick start example is available at beaver/sample_data/. To play with this example:

./beaver ../sample_data/sample_input.list ../sample_data/beaver_test > test.log

Output will be generated at ./sample_data/ with prefix beaver_test.

Format of Input and Output

Input Individual Assembly List

Each line of input-gtf-list points a path to the individual assembly (GTF format) of a single cell or sample; see sample_input.list.

Output Files Structure

1. Cell-Specific Assemblies
   • Location: output-prefix_sgtf/X.gtf
   • X ranges from 1 to number of cells
2. Meta Assembly
   • Location: output-prefix.gtf
   • Contains collective transcripts from all cells
3. Feature Files
   • General features (output-prefix_feature.csv)
   • Cell-specific features (output-prefix_sgtf/X_feature.csv); X ranges from 1 to number of cells

Options

Parameters	Type	Default Value	Description
-t	integer	1	Number of threads.
-c	double	0	Minimum transcript output coverage (0-1)
-pn	integer	15	Maximum number of paths per node.
-pg	integer	100	Maximum number of paths per graph.

Scoring Cell-Specific Transcripts

Beaver employs a two-stage Random Forest pipeline (Beaver_General and Beaver_Specific) for scoring transcripts, which is available for download from Zenodo. Use the provided Python scripts Beaver_General.py and Beaver_Specific.py with the pre-trained models. Please note that if you only care about the global expression of cell populations in the dataset, Beaver_General can work independently without Beaver_Specific.

Dependencies

Python version tested: Python 3.11.7 Required Python libraries: numPy, pandas, scikit-learn, joblib

• Using pip:

  pip install numpy pandas scikit-learn joblib

• Using conda (recommended):

  conda install numpy pandas scikit-learn joblib

Beaver_General Usage

python3 beaver_general.py -n <num_cells> -i <general_transcript_feature_csv> -m <pretrained_general_model.joblib> -o <general_features>

Parameter	Type	Default	Description
-i	String		Input general feature CSV file; e.g. {output-prefix}_feature.csv.
-m	String		Path to the pre-trained general model file for scoring.
-n	Integer		Number of cells/samples.
-o	String		Output Beaver-General features CSV file.
-p	Float	0.2	Minimum probability score threshold (range: 0 to 1).

Beaver_Specific Usage

python3 beaver_specific.py -n <num_cells> -i <specific_transcript_feature_dir> -g <general_features> -m <pretrained_specific_model.joblib> -o <output_specific_dir>

Parameter	Type	Default	Description
-i	String		Input cell-specific feature directory.
-m	String		Path to the pre-trained specific model file for scoring.
-n	Integer		Number of cells/samples.
-g	String		Beaver-General features CSV file.
-o	String		Output directory.
-p	String	0.6	Minimum specific probability score threshold (range: 0 to 1).
-pg	String	0.2	Minimum general probability score threshold (range: 0 to 1).
-s	-		Save processed specific feature files in the output directory.

Example

python3 beaver_general.py -n 2 -i sample_data/beaver_test_feature.csv -m model_real_HEK293T_meta_roc=0.824.joblib -p 0.2 -o sample_data/beaver_general_scores.csv

python3 beaver_specific.py -s -n 2 -i sample_data/beaver_test_sgtf -g sample_data/beaver_general_scores.csv -m model_real_HEK293T_specific_roc=0.796.joblib -p 0.8 -pg 0.2 -o sample_data/beaver_test_specific