Awesome

How to build BENCHFAM !

BenchFam creates automatically a protein benchmark dataset that can be used to evaluate the performance of alignment programs, even for large scale aligners.

This is done by getting all protein families from the PFAM database that belong that contain at least 10 members with known crystallographic structures. This treshold can be modify in an early stage when preparing the data to give to the nextflow workflow. The quality of the stuctures depends on the results of the BLAST against the PDB database (it may not always be the best).

1 - Installing NextFlow and benchfam

NextFlow

Install Nextflow runtime with this command:

curl -fsSL get.nextflow.io | bash

Prerequisites

• Java 7 or later
• Docker engine 1.0 or later (in alternative you can install the required dependecies as shown in the included Dockerfile)

BenchFam

BenchFam is on GitHub; clone it from the CBCRG git account:

git clone https://github.com/cbcrg/benchfam.git

All PERL and bash scripts, binaries, etc...are in the $CWD/benchfam/bin/.

2 - Getting the data

Download the PFAM database

You can either download from the webpage directly or by using wget:

wget ftp://ftp.ebi.ac.uk/pub/databases/Pfam/releases/Pfam28.0/Pfam-A.full.gz

Uncompress the PFAM database

It is a big single text file (>100G), it takes time !!!

gunzip Pfam-A.full.gz

Extract sequences and alignment from PFAM

The PFAM database is a single Stockholm formatted text file. It is therefore necessary to extract all the families beforehand. To do so, use the perl script PFAM_extract.pl. It requires three argument:

• "FULL" or "PDB" (full extract all sequences from a family; PDB only 3D)
• number (number max of family to extract; anything except a number extract all)
• the Pfam database file (or any Pfam formatted file)

The script will extract in the current working directory all the families of PFAM aligned and unaligned; you have to run it twice (two terminals) to get all sequences or only PDB ones:

perl PFAM_extract.pl PDB X Pfam-A.full
perl PFAM_extract.pl FULL X Pfam-A.full

CAUTION: this step takes ~1 day, because of the size of the database

Create a working directory for NextFlow workflow

You must create a folder containing families with a minimum of 10 sequences that will be analyze by the nextflow pipeline. To do this use the following bash script:

./PFAM_count.sh

For the moment the script must be located where the files are, but this can be changed. It is an interactive script that will ask you two parameters:

• X: the minimum number of sequences per file(10)
• folder where to move the families with Nseq >= X sequences
• folder where to move the families with Nseq < X sequences

3 - Run BenchFam with NextFlow

There are still some improvement to do; for now, the path of the files extracted from Pfam-A.full has to be harcoded inside the nexflow piepline (benchfam.nf). First, you need to get BenchFam from github and have NextFlow installed (step 1).

Description of benchfam.nf

The BenchFam pipeline performs 3 operations:

• 1_filter_PDB:

Performs a first layer of filtering of the number of sequences, %id, coverage, etc...and identifies 3D template from the PDB.

• 2_extract_PDB:

Performs an extract of the exact part of the structure file corresponding to the sequences within the sequence file. It cuts out the 3D structure of each protein and rename the file with an index according to the number of domains within the same PDB file (it is important especially when domains are repeats).

• 3_align_libraries:

Performs all alignments using a multitude of aligners based on structures or on sequence (e.g. SAP, TMalign, T-Coffee...)

Running benchfam.nf

The command line to run benchfam.nf is:

nextflow run benchfam.nf -c pfam28_2018.config

If the run is stopped for any reason, it can be resumed:

nextflow run benchfam.nf -c pfam28_2018.config -resume

4 - Organize & Test BenchFam output

The BenchFam pipeline will generate many output files stored in a scratch folder even when for some datasets fail at some step of the workflow. To get & organize the data, we use a perl script : BENCHFAM_LOG_extract.pl. The script needs two input arguments:

• the log file from BenchFam nextFlow
• the full PATH for the scratch directory

Generate the script to organize/test BenchFam:

perl BENCHFAM_LOG_extract.pl trace_pfam_28.0_2018 $PATH/PFAM_28.0_scratch

The script generated before is a bash script run_copy_scratch.sh which has to be run in the SCRATCH folder $PATH/PFAM_28.0_scratch in order to organize BENCHFAM:

./run_copy_scratch.sh

The tests will concern only eventual discrepancies between the PDB information from BLAST and the information in the corresponding PDB file (missing regions, mutations, etc...). The evaluation is made one the IRSMD files from the EVAL folder; any structural discrepancy will results in a IRMSD value of -1.00. All families for which this happen will be moved in a folder "PROBLEMS".