DEEPEST (Data-Enriched Efficient PrEcise STatistical Fusion detection) is a statistical fusion detection algorithm developed in the Salzman Lab at Stanford University. DEEPEST also contains a suite of pragrams for detecting linear and circular junctions in addition to fusion junctions. There are two options for running DEEPEST: (1) running DEEPEST online tool with only few clicks on the CGC platform; or (2) running DEEPEST workflow on a local high performance computing cluster via downloading the provided wrapper and index files. The wrapper is compatible with any local cluster environment and does not need any software installation in advance.
An online tool for DEEPEST is now publicly available on Cancer Genomics Cloud (CGC) at: https://cgc.sbgenomics.com/public/apps#jordanski.milos/deepest-fusion/deepest-fusion/
The tool is ready-to-run with only few clicks and can be run either on the dataset uploaded by the user or on public RNA-Seq datasets (such as TCGA, TARGET, CCLE, ...) already available on CGC.
To run the online tool, you need to login to CGG here. Each new account would automatically have $300 pilot funds. Our current cost estimate for running DEEPEST on a typical RNA-Seq dataset (such as TCGA or GTEx) is $4-$5 per sample.
The following software prgrams are needed to run DEEPEST on a local cluster. They just need to be available as modules on the cluster and DEEPEST automatically loads them with module load command.
- Anaconda
- SciPy 1.1.0
- Bowtie2 2.2.9
- Python 3.4 with Biopython/1.70 installed
- java
- R 3.5.1
- Trim Galore! 0.4.4
All computational steps including any alignment step required for running DEEPEST has been wrapped in a single JSON file DEEPEST-Fusion_pipeline.json. This script can be run on any local cluster using an input JSON file that provides the paths for reference index files and input RNA-seq files. To run the script, Rabix toolkit should be pre-installed on the local cluster.
The paths to input files required for running DEEPEST should be provided via a single input JSON file DEEPEST-Fusion_input.json. The following parameters should be set in the input JSON file:
- Bowtie2 index files for the reference genome
- Bowtie2 index files for the regular junctions
- Bowtie2 index files for the scrambled junctions
- Bowtie2 index files for ribosome
- Bowtie2 index files for the transcriptome
- Bowtie2 index files for indel junctions (junctions with up to 5 symmetric indels at their splice sites)
- Fastq files for the input RNA-Seq data
- Pickle files for genes/exons annotation
- Pickle file for the known fusions list (a database of candidate fusions constructed based on
ChimerDB 3.0, a curated list of known cancer fusions) - Bowtie2 index files for known fusions
You can download preprocessed and ready-to-use index files from: https://drive.google.com/open?id=1ftT5UjKafcsDzUFlV49szOD8SiH3-Qws and provide the paths to the location they have been downloaded in DEEPEST-Fusion_input.json.
You can also build the index files yourself. The scripts/instructions needed for building the index files along with the reference fasta files (genome, transcriptome, regular junctions, scrambled junctions, ribosome) are provided here: https://github.com/salzmanlab/DEEPEST-Fusion/tree/master/reference_files
For running DEEPEST on a local cluster, Rabix should be installed first. Rabix is an open-source tool developed by Seven Bridges, that can be used to run a computational workflow written in Common Workflow Language (CWL) on a locul cluster. More information on how to install Rabix can be found in this GitHub repositiory: https://github.com/rabix/bunny
An example batch script DEEPEST-Fusion_submit_job.sbatch, based on job scheduler Slurm has been provided. In the batch script file the path to where Rabix has been installed, DEEPEST pipeline JSON file (DEEPEST-Fusion_pipeline.json), and DEEPEST input JSON file (DEEPEST-Fusion_input.json) should be provided.
Three primary report files containing reported fusion junction with their corresponding statsitical scores and number of various types of aligned reads can be found as follows:
- modified-MACHETE report file (based on FarJunctions database):
Knife_and_MACHETE_Known_fusions_parallel_rev_*/root/MACHETE_AppendNaiveReptParallel/[Dataset name]_naive_report_AppendedMACHETE_Parallel.txt - modified-MACHETE report file (based on known fusion database):
Knife_and_MACHETE_Known_fusions_parallel_rev_*/root/MACHETE_AppendNaiveReptParallel_Known/[Dataset name]_naive_report_Appended_MACHETE_Parallel_Known.txt - KNIFE report file:
Knife_and_MACHETE_Known_fusions_parallel_rev_*/root/KNIFE_GLM_model/[Dataset name]_1_circJuncProbs.txt_cdf
In each report file, high confidence junctions are called via filtering based on the statistical scores (all needed filters and their thresholds are described in the paper). A list of nominated fusion junctions as the output of the junction nomination component (the first component in DEEPEST) is reported as the union of fusion junctions called from the above three report files. This list of nominated fusions by the first component needs to undergo a statistical refinement step, which is based on Sequence Bloom Trees.
All scripts needed for implementing Sequence Bloom Tree (SBT) filters for an RNA-Seq database can be found in this github repository: https://github.com/elehnertSBG/SBT-Apps. Mor information on the SBT algorithm and the order in which the SBT CWL scripts should be run can be found in the original SBT manual: https://www.cs.cmu.edu/~ckingsf/software/bloomtree/sbt-manual.pdf
Dehghannasiri, R., Freeman, D.E., Jordanski, M., Hsieh, G.L., Damljanovic, A., Lehnert, E. and Salzman, J. "Improved detection of gene fusions by applying statistical methods reveals oncogenic RNA cancer drivers", Proceedings of the National Academy of Sciences, July 2019. (https://www.pnas.org/content/116/31/15524)
In case of any questions, please contact Roozbeh Dehghannasiri (rdehghan@stanford.edu) or Milos Jordanski (milos.jordanski@stanford.edu).