Skip to content

Latest commit

 

History

History
114 lines (83 loc) · 7.34 KB

README.md

File metadata and controls

114 lines (83 loc) · 7.34 KB
Raw

Overview

This repository is dedicated to comparing the performance of the meta-assembler Aletsch against two leading meta-assemblers, TransMeta, PsiCLASS, as well as two meta-assembly pipelines based on single-sample assemblers, StringTie2-Merge and Scallop2 combined with TACO. Here we provide instructions for downloading necessary tools, preparing datasets, executing the tools/pipelines, scoring Aletsch's output transcripts, and reproducing the results presented in the Aletsch paper.

Step 1: Download and Link Tools

Our experiments involve the following tools:

Tool Version Description
Aletsch v1.1.0 Meta Assembler
Transmeta v.1.0 Meta Assembler
PsiCLASS v1.0.3 Meta Assembler
StringTie2 v2.2.1 Single-sample Assembler + GTF merged tool
Scallop2 v1.1.2 Single-sample Assembler
TACO v0.7.3 GTF merged tool
STAR v2.7.11 RNA-seq Aligner
GffCompare v0.11.2 Evaluate assembled transcripts
gtfcuff - RNA-seq tool

Step 1.1: Download Tools

  • Access the homepages of the respective tools using the links provided above.
  • Follow the download and compilation instructions on each tool's homepage.

Step 1.2: Link or Copy Executables

  • For tools with available executable files, link or copy them to the programs directory. This includes aletsch, scallop2, taco_run, stringtie, STAR, gffcompare, and gtfcuff.
  • For tools without standalone executables (TransMeta and PsiCLASS), link the entire directory to programs.

Ensure the tools are accessible via the following paths:

your/path/to/programs/aletsch
your/path/to/programs/TransMeta/TransMeta
your/path/to/programs/psiclass/psiclass
your/path/to/programs/stringtie
your/path/to/programs/scallop2
your/path/to/programs/taco_run
your/path/to/programs/STAR
your/path/to/programs/gffcomapre
your/path/to/programs/gtfcuff

You may need to rename some of the executable files to match the paths listed above.

Step 2: Download Datasets and Align

We evaluate the performance of the five methods using eight datasets, as outlined below. Each dataset is identified by its unique prefix (used in this repository) and accession ID for reference.

Name in paper prefix in test(Ensembl) Protocol Accession ID
BK-H1 encode10_ensembl Illumina paired-end Refer to data/encode10.sra.list
BK-H2 PRJNA575230_ensembl Illumina paired-end Refer to data/PRJNA575230.sra.list
SC-H1 smartseq3_ensembl_human_100 Smartseq3 Random 100 cells from HEK293T of E-MTAB-8735
SC-H2 smartseq3_ensembl_xpress_run2 Smartseq3-Xpress All 1066 cells from PBMCs_run2 of E-MTAB-11452
BK-H3 PRJNA489891_ensembl Illumina paired-end Refer to data/PRJNA489891.sra.list
SC-H3 smartseq3_ensembl_human_92 Smartseq3 Random 92 cells from HEK293T of E-MTAB-8735
BK-M1 PRJEB18790_ensembl Illumina paired-end Refer to data/PRJEB18790.sra.list
SC-M1 smartseq3_ensembl_mouse Smartseq3 All 369 cells from Mouse-Fibroblast of E-MTAB-8735

Use STAR for read alignments for each sample/cell. For every dataset, compile a list of all BAM file paths as required by the different meta-assemblers. Example for Aletsch: data/encode10_ensembl.star.list. Your lists should follow this format:

your/path/to/data/encode10_ensembl.star.list
your/path/to/data/PRJNA575230_ensembl.star.list
...

Annotations for evaluation are available from Ensembl and RefSeq.

Step 3: Run the Methods

Execute the provided scripts in the results directory to run the assemblers:

./run.aletsch.sh 
./run.transmeta.sh
./run.psiclass.sh
./run.stringtie2.sh
./run.st2merge.sh
./run.scallop.sh
./run.sc2taco.sh

Each script accepts different list arguments for the assembler. Example usage:

Usage: ./run.aletsch.sh {encode10|xx230|sc-human|sc-xpress|xx891|mouse790|sc-mouse} dataset_index
For example: ./run.aletsch.sh xx230 3

Note: Ensure run.stringtie2.sh is completed before run.st2merge.sh, and run.scallop.sh before run.sc2taco.sh. Scripts may be adjusted for parallel processing with specified CPU core usage. BAM files for specific chromosomes are prepared using samtools, e.g., for extracting chr1:

samtools view -b -F 4 -@ 10 -o "$chr1_bam" "$bam_file" 1

Alternatively, assemble with all chromosomes and later extract chr1 results from the GTF file.

Step 4: Test with Aletsch's Pretrained Model and Compare

Navigate to the random-forest directory and run prepare.sh as done in Step 3, e.g., ./prepare.sh xx891 2. This will collect Aletsch's feature files in your/path/to/random-forest/aletsch.

Download Aletsch's pretrained model from Zenodo, typically "Aletsch-ChrAll.ensembl.joblib". Execute the following command to test and compare the results:

cd your/path/to/random-forest
./prepare.sh xx891 2 # Example for dataset "PRJNA489891_ensembl"
python3 test.py -i xx891

Ensure you have the necessary Python packages (numpy, pandas, sklearn) installed. The test.py script can be modified to use a different pre-trained model, with outputs stored in your/path/to/random-forest/logs. An example log file, Aletsch-Chr1.ensembl.sc-human100.other_chrs.log, demonstrates the result of Aletsch-Chr1 on dataset SC-H1, including metrics like adjusted precision and pAUC for comparison.