shao-group / aletsch

Assembler for multiple RNA-seq samples

Introduction

Aletsch implements an efficient algorithm to assemble multiple RNA-seq samples (or multiple cells for single-cell RNA-seq data). The datasets and scripts used to compare the performance of Aletsch with other assemblers are available at aletsch-test.

Version v1.1.1

We released Aletsch v1.1.1, a version that substantially improved the memory usage and running time over its previous version v1.1.0, while maintaining an identical assembly accuracy. The improvement was primarily made by fixing the incorrect use of bam-file queries and by removing PCR duplicates. Below we detail the memory usage and running time, both CPU-time and Wall-time (10 threads), of the two versions across all datasets we tested (see aletsch-test).

Memory Usage Comparison (in GB):

Dataset	v1.1.1	v1.1.0
BK-H1	6.96	35.55
BK-H2	11.79	64.44
BK-H3	5.32	34.35
BK-M1	21.47	168.01
SC-H1&3	4.12	47.23
SC-H2	24.43	251.81
SC-M1	9.27	82.93

CPU And Wall-Clock Time Comparison (in minutes):

Dataset	v1.1.1(CPU)	v1.1.1(Wall)	v1.1.0(CPU)	v1.1.0(Wall)
BK-H1	219	27	541	53
BK-H2	923	96	1319	135
BK-H3	155	17	258	28
BK-M1	691	73	1464	169
SC-H1&3	186	21	167	20
SC-H2	1077	129	1530	183
SC-M1	382	44	441	52

Installation

Aletsch can be installed through conda or by compiling source (see INSTALLATION).

Usage

The usage of aletsch is:

./aletsch -i <input-bam-list> -o <output.gtf> [options]

We highly recommend to generate profiles for individual samples first:

./aletsch --profile -i <input-bam-list> -p <profile>
./aletsch -i <input-bam-list> -o <output.gtf> -p <profile> [options]

Format of Input and Output

Each line of input-bam-list describes a single sample, with 3 fields separated by space. The 3 fields are: alignment-file (in .bam format), index-alignment-file (in. bai format), and protocol. The index-file can be generated using samtools (e.g., samtools index ...). The protocol is chosen from the 5 options: single_end (for illumina single-end RNA-seq protocol), paired_end (for illumina paired-end RNA-seq protocol), pacbio_ccs (for PacBio Iso-Seq CCS reads), pacbio_sub (for PacBio Iso-Seq sub-reads), ont (for Oxford Nanopore RNA-seq). Aletsch will use different parameters / algorithms to process different data types.

Aletsch requires that each input alignment file is sorted; otherwise run samtools to sort it (samtools sort input.bam > input.sort.bam).

The assembled transcripts from all these samples will be written to output.gtf, in standard .gtf format.

Options

Aletsch provides several options for transcript assembly, supporting both its unique parameters and those required by the core algorithm of Scallop. For a detailed list, execute aletsch without arguments.

Parameters	Type	Default Value	Description
--help			Displays Aletsch usage information and exits.
--version			Shows Aletsch version information and exits.
--profile			Profiles individual samples and exits. Writes to files if -p is specified.
-l	string		Specifies chromosomes to assemble.
-L	string		Specifies a file containing a list of chromosomes to assemble.
-d	string		Output directory for individual sample transcripts. Directory must exist prior to execution.
-p	string		Directory for reading/saving individual sample profiles. Directory must exist prior to execution.
-t	integer	10	Number of threads.
-c	integer	200	Maximum number of splice graphs in a cluster, recommended as twice the number of samples.
-s	float	0.2	Minimum similarity for combining two splice graphs.

• If -l string or -L file option is provided, Aletsch assembles only the specified chromosomes; otherwise, it assembles all chromosomes.

• Directories specified by -d and -p must exist before running Aletsch; the tool does not create directories.
• With --profile, Aletsch infers profiles of individual samples, using the XS tag from input BAM files.

Scoring Transcripts with Pre-trained Model

Aletsch employs a random forest model for scoring transcripts, available for download from Zenodo. Use the provided Python script score.py with this model.

Dependencies

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

Usage

Score transcripts with the syntax below:

python3 score.py -i <individual_gtf_dir> -m <pretrained_model.joblib> -c <num_of_samples> -p <min_probability_score> -o <output_score.csv>

Parameter	Type	Default	Description
-i	String		Directory containing Aletsch's feature files(x.trstFeature.csv). This is the same directory where Aletsch outputs individual GTF files, as designated by the -d option in Aletsch's assembly process.
-m	String		Path to the pre-trained model file for scoring.
-c	Integer		Number of samples/cells
-p	String	0.2	Minimum probability score threshold (range: 0 to 1).
-o	String		Output directory of scored .csv file.

Assuming a collection of $n$ samples, the directory <individual_gtf_dir> contains a total of $n+1$ feature files, enumerated from 0.trstFeature.csv through to n.trstFeature.csv. Files 0.trstFeature.csv to (n-1).trstFeature.csv correspond to feature files for individual samples, sequentially from the first to the last sample. The file n.trstFeature.csv is derived from the combined graph.

aletsch's Issues

There are many duplicate instance names in aletsch result file

Hi,

There are many duplicate instance id in the results file, is there a way to avoid?

dup.instance.zip

Best,
Kun

Which version of scallop was employed in aletsch?

Hi,

Which version of scallop was employed in aletsch, scallop or scallop2 ?

Best,
Kun

Can I use aletsch to assemble single bulk rnaseq sample?

Hi,

I found that aletsch is much faster than scallop2, I want to known if I can use it to assemble single bulk-rnaseq sample (or multiple bulk-rnaseq sample merged bam) instead of using scallop2 ?
Why or why not?

Best,
Kun