Cancer Systems Biology, Technical University of Denmark, 2800, Lyngby, Denmark & Cancer Structural Biology, Danish Cancer Institute, 2100, Copenhagen, Denmark
Repository associated with the Preprint:
PDBminer to Find and Annotate Protein Structures for Computational Analysis
Kristine Degn, Ludovica Beltrame, Matteo Tiberti, Elena Papaleo
bioRxiv 2023.05.06.539447; doi: https://doi.org/10.1101/2023.05.06.539447
PDBminer is a program generating a ranked overview of the available structural models in the Protein Data Bank and the most current version of the AlphaFold2 model for an input protein using the UniProt accession number. PDBminer captures existing structural model details, including chains associated with the UniProt accession number, model quality (e.g., resolution and r-free), re-aligned coverage of the model towards the UniProt sequence (of any isoform) excluding missing residues, protein complex or fusion products within the structure file as well as the presence of nucleic acid chains or any ligands, and if a PDB-REDO refined structure exists. Additionally, the pLDDT score or b-factor is reported. Hence, the output table contains a wide range of information suitable for deciding on a structural model for further research.
- Installation
- Setup
- Running PDBminer
- Understanding the Output
- Plotting the Output
It is recommended to create a virtual environment to run PDBminer. The environment can be installed via conda using the environment.yml or via pip with requirements.txt.
PDBminer requires:
- python3
- biopython
- numpy
- pandas
- PyYAML
- requests
and for plotting:
- seaborn
- matplotlib
- networkx
The versions of these dependencies are available in requirements.txt and environment.yml. PDBminer is developed on a Linux based system but also tested on MacOS and Windows. Notice that for BioPython, it is important to use version 1.78.
git clone https://github.com/ELELAB/PDBminer.git
cd PDBminer
python3 -m venv PDBminer_env
source PDBminer_env/bin/activate
python3 -m pip --default-timeout=1000 install -r requirements.txt
source PDBminer_env/bin/activate
git clone https://github.com/ELELAB/PDBminer.git
cd PDBminer
conda env create -f environment.yml
conda activate PDBminer_env
conda activate PDBminer_env
There are two ways of running PDBminer. Either by using an input file, or by using the command line to find the available structures for a single protein. In the directory examples are three examples and their commands in the do.sh file. Consider testing the installation and use by running one or more of these.
The input file should be in a comma-separated values file format, with one mandatory column; "uniprot" containing the accession number of the uniprot entry of interest. Additionally the optional columns are "hugo_name", "uniprot_isoform", "mutations" and "cluster_id".
Example of input file as a table:
hugo_name,uniprot,uniprot_isoform,mutations,cluster_id
MAT1A,Q00266,1,P30N;W300H,1
SSTR3,P05543,1,T11S;C191S;R330L,1
SAMD4A,Q9UPU9,3,L10R;I80A,1
TP53,P04637,2,P278L;R337C;L344P,1
The minimal required content for the input file:
uniprot
Q00266
P05543
Q9UPU9
P04637
The name of the input file should be specified in the command line.
- "hugo_name": The gene name can, in principle, be anything and can also be used to assign a run-specific name relevant to the user.
- "uniprot_isoform": The isoform reflects the sequence PDBminer aligns the sequence of the structures to and assigned mismatches between the structure sequence and the UniProt sequence.
- "mutations": If you input mutations, PDBminer will filter the structures based on the sites of the mutations. That means every structure in the filtered output covers at least one mutational site. It does not mean that the mutation necessarily is present in the structure. Using mutations is a way to limit the search space to areas of interest. If mutations are part of the input, a column in the output will indicate the amino acids in the structure sequence at the mutational site.
- "cluster_id": If you have multiple mutations that you already know may be in different domains of the protein, it may be beneficial to use the cluster ID because you parse the filtered file into more sections, covering different domains.
$ python PDBminer -i [input file name] -n [cores] -f [output_format]
When PDBminer is run on only a single protein it may sometimes be beneficial to run it directly in the commandline. To do so, a input file does not need to be constructured and the content can be specified with flags. Again, the uniprot option is mandatory while the rest is optional.
$ python PDBminer -g [hugo_name] -u [uniprot_id] -s [uniprot_isoform] -m [mutations] -c [cluster_id] -n [cores] -f [output_format]
$ python PDBminer -g SSTR3 -u P05543 -m "T11S;C191S;R330L" -n 1 -f json
$ python PDBminer -u P05543
NOTICE: when isoform is not specified 1 is assumed. NOTICE: json is the default output format, csv can be chosen (but is discouraged).
The output is a log.txt file, a input_file.csv and a directly "results". A log.txt file is created for each run and contains information regarding the run, e.g. if the input is false or there are any connectivity issues.
In the directory "results", a subdirectory for each uniprot accession number is created. After a successful run, the uniprot accession number directory can contain the following:
-
log.txt, If there are no structures from the protein data bank or alphafold structures available for the uniprot_id the log.txt file is the only output. If there are any errors or warnings while running a particular protein, these are also listed in the log.txt file. This can be used for error handling.
-
{unipot_id}_all.json (or .csv if the option is chosen). An output file with all PDBs and AlphaFold structure associated with the uniprot_id regardless of mutational coverage. Notice that you may validate the json file towards the schema.json.
If mutations are included in the input, a filtered version of all will also be available if the mutations are covered by any structure.
- {unipot_id}_filtered.csv or .json, An output file with the PDBs and alphafold structure associated with the uniprot_id that covers at least one mutation.
Notice that multiple filtered files are available when multiple clusters are parsed.
- {uniprot_id}_cluster{cluster_id}_filtered.json/csv.
See examples of the in- and output of the example directories.
#content of {unipot_id}_filtered.json/csv and {unipot_id}_all.json/csv:
- structure_rank: Index, the lower the value the seemingly better the model.
- hugo_name: Gene name from the input.
- uniprot_id: Uniprot id from the input.
- uniprot_isoform: Uniprot isoform from the input or 1 if none given.
- mutations: A list of input mutations, only visible in the filtered output.
- cluster_id: If clusters are specified, the cluster will be visible in this column.
- structure_id: Identifier of the PDB file or Alphafold Model.
- deposition_date: Timing of file placement in PDB or model generation in the Alphafold Database.
- experimental_method: By which the PDB was generated. For AlphaFold model "predicted" is used.
- resolution: Estimation of PDB quality (for X-ray structures).
- r-free: the r-free value for the structure.
- PDBREDOdb: A YES/NO column if the structure is available in the PDB-REDO database.
- PDBREDOdb_rfree: The r-free of the PDB-REDO refined structure.
- complex_protein: A column defining if a complex or fusion is present in the PDB file.
- complex_protein_details: Details regarding the protein complex indicating the Uniprot ID of the other protein and the chains.
- complex_nucleotide: Binary, indicates if the protein is bound to a nucleotide string such as DNA.
- complex_nucleotide_details: Details regarding the DNA or RNA binding.
- complex_ligand: Binary indicating if metal or small molecules are present in the pdb file.
- complex_ligand_details: Details describing which “other” things are in the file.
- chains: Letter or letters describing the chains covering the Uniprot ID
- coverage: Coverage is a range of numbers indicating the area the model covers the uniprot sequence using the uniprot numbering of the sequence. One chain can have multiple sequences, for PDB structures indicating missing residues and for AlphaFold structures when the pLDDT score is below 70.
- mutations_in_pdb: When structures contain amino acid sequence that differ from the sequence specified with the isoform, they are annotated as mutations and can be found here.
- b_factor: A dictionary of the b-factor of the chains.
- warnings: This coloumn contains information regarding dissimilarities or discrepancies that cannot be explained by fusions or other known alterations. This includes expression tags and amino acids added to terminals for structure solving purposes.
For all columns ";" seperate data on the annotated chains and "NA" indicates that no relevant data is present.
PDBminer2coverage is a plotting tool creating an overview of the protein sequence on the x-axis and the different models covering the sequence on the y-axis. The area the model covers is colored grey. If any positions are mutated, the position will be colored in with a transparent blue hue across all entries on the y-axis. PDBminer2coverage takes the results directory and input file as required input, per default these are input_file.csv and the current working directory/results. Hence, the PDBminer2coverage module can be run in the same place as PDBminer without any arguments. The output is one or more plots. If there are both a filtered- and an all-output file, both will be plotted. If there are multiple clusters, these will be plotted separately. In cases where the protein sequence is longer than 500 amino acids, the plot will be split into multiple output files, termed "chunks". Additionally, it is possible to narrow down the plotting area with the flag -s --sequence. Mutations in the PDB files are colored red.
$ PDBminer2coverage -s 1-20,50-95
Would, for example, only plot the sequence 1-20 and 50-95 in the same plot.
Additionally you can also set a limit on the x-axis, indicating how many positions you want plotted using -t.
$ PDBminer2coverage -t 50
would, for example only plot 50 amino acids per chunk. Default is 500.
Flags:
-r: choosing the results path if not default.
-i: The input file.
-u: uniprot id can be added if only one of the proteins in a multi protein run should be visualized.
-s: sequence, dash and comma separated values such as 1-10 or 1-20,30-40 for the sequence of the protein to plot.
-t: threshold, a integer of the sequence length to be placed in each individual plot, default is 500, why each plot is maximum 500 amino acid broad.
-bb: The value the best b-factors are below, integer.
-bg: The value good b-factors are below, integer.
-bp: The value poor b-factors are above, integer.
All options, example:
$ PDBminer2coverage -r PDBminer_run/results -i PDBminer_run/input_file.csv -u P00000 -s 30-120 -t 100 -bb 10 -bg 20 -bp 30
PDBminer2network is a plotting tool creating an overview of the protein complexes within the protein data bank for the protein of interest. PDBminer2network takes the current working directory/results directory as default input, just as the PDBminer2coverage module. The output is a network with the Uniprot ID of your protein at the center, branching out to "protein complexes" and/or "fusion products" depending on the content of the output file. From here each bound or fused protein is the first node and the second is the PDBid with the complex or fusion. The output is one or more plots. If there are both a filtered- and an all-output file, both will be plotted. If there are multiple clusters, they will be plotted seperately.
$ PDBminer2network -h
Would, for example, write out the help information.
The plot can be adjusted using the following flags:
-r: choosing the results path if not default.
-i: The input file.
-u: uniprot id can be added if only one of the proteins in a multi protein run should be visualized.
-c: node color for center of graph
-p: node color for proteins (fused and bound)
-s: node color for structures (PDBid)
-t: color for the nodes with "protein complex" and "fusion product".
The edges of the graph are black. Any color can be used e.g. '#183233’.
All options, example:
$ PDBminer2network -r PDBminer_run/results -i PDBminer_run/input_file.csv -u P00000 -c '#64b2b5' -p '#183233' -s '#1fc6cc' -t '#a3cacc'
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent