Biological Data

This goal of this project is to investigate the functional and structural properties of the SH2 domain starting from a sample sequence. Our sequence is identified by UniProt ID P23615(1258-1339) and can be found as a fasta file data/sequenceP23615.fasta.

Requirements

The project have been developed using Python 3 with the packages cotained in requirements.txt. It is possible to install the dependencies using pip:

 pip install -r requirements.txt

For the structural alignment the softwar TMalign is needed. Source code and instruction on how to compile it can be found in code/part2/TMalign. The executable present in the directory has been compiled on OSX.

Moreover BLAST and HMMER are required and should be accessible from PATH.

We used jalview to edit the multiple sequence alignments.

Part 1: Domain Models

The goal of this first part is to build a PSSM and HMM models representing the asigned domain.

Step 1

The first step is retrieve homologous sequences from UniProt. To do this we used Blast on UniRef90 and 500 hits. These sequences are save in a fasta file data/BLAST_uniref90.fasta.

Step 2

The retrieved hits have been used to generate a multiple sequence alignment usign Clustal Omega. The aligment is saved in data/msa_clustalw.fasta. We then edited the MSA usign JalView. The result is saved in data/msa_edited.fasta

Step 3

In this step we created and evaluated the models. Models are evaluated against Pfam annotation of the domain, PF00017, in the human organism.

To do this we need to get from UniProt all the human sequences available in SwissProt containing our domain. This can be done with the query on UniProt

organism:"Homo sapiens (Human) [9606] AND reviewed:yes

Results have been downloaded and saved in data/SwissProt_humans_reference_all.fasta. The list of true positive, i.e. sequences containing our domain, can be retrieved with a similar query

database:(type:pfam pf00017) AND reviewed:yes AND organism:"Homo sapiens (Human) [9606]

This has been saved as data/SwissProt_humans_reference.fasta, even if it is not necessary.

These files can be used to crete a BLAST database (Blast indexes, phr+pin+psq files) that will be used by HMM-SEARCH and PSI-BLAST to retrieve proteins:

makeblastdb -dbtype prot -in data/SwissProt_humans_reference_all.fasta -parse_seqids

You can test if it is working correctly by searching a sequence in the database:

blastdbcmd -entry "CRK_HUMAN" -db data/SwissProt_humans_reference_all.fasta

Creation and evaluation of the models are described on the notebook code/part1/ModelsEvaluation.ipynb. Here you can find also a comparison between them. Scripts performing all the steps automatically are in code/part1/{psiblast.py, hmm.py, jackhmmer.py}. Other scripts in code/part1/ contain utility functions. To run one of them use

python code/part1/psiblast.py

Created models are saved in the directory models and search output on results. Parameters for this scripts are contained in the headers.

Part 2: Domain family characterization

Once the list of human sequences matching the model is defined, we can look at functional and structural properties.

The first step consists in the creation of four different datasets that will be enriched (gene ontology and disease ontology).

The first one is called Original dataset and is composed of the protein retrieved by our model against human proteins in SwissProt. It can be found in datasets/original.txt and it is created by the scripts described in part 1.

The second is called Architectures datasets. This dataset is composed by multiple datasets, where each of them contains proteins of the original dataset with the same combination of Pfam domains. The script used to create this dataset is code/part2/architectures_dataset.py and it saves the dataset in datasets/architectures_datasets.json. Domains are used as key of the json.

The third is called PDB_network and it contains all the sequences of the original dataset with a PDB plus other human proteins found as chains in the same PDB. The dataset is created with the script code/part2/pdb_network.py and saved in datasets/pdb.csv. To find the PDBs the script utilize the file data/pdb_chain_uniprot.tsv.gz that can be downloaded from SIFTS Together with the sequence and PDB id we saved the position of the PDB in the sequences. This will be used later on to find PDBs covering our domain.

The last dataset is called STRING network, composed by all the proteins in the original dataset plus all direct interactors found in the STRING database. This dataset is created with the script code/part2/string_network.py and save in datasets/string.txt.

In the notebooks code/part2/{GeneOntology.ipynb, DiseaseOntology.ipynb} the background datasets are identified and the enrichment is performed.

Structural classification

The last part is related to the structural classification. This is composed by two sections: one where we provided statistics about CATH architectures mapping to our domain and a second where the structural alignment of PDBs covering our domain in performed. Procedure and Results relative to CATH are presented in the notebook that can be found in code/part2/CATH_statistics.ipynb.

For the second part, PDBs covering our domain are found with the script code/part2/findCoverignPDBs.py and downloaded in datasets/coverign_pdbs. A PDB is identified as covering the domain if the overlap with the predicted position has an F1 greather than a certain threshold (default value is 0.7). By changing the threshold the list of pdbs will change and the script look if these files have already been downloaded, otherwise it will get them from PDB. These files are used to performed an all-vs-all pairwise structural alignment. The scores are used to build a distance matrix and a dendrogram representing the hierarchical clustering is created. This procedure and results are presented in the notebook code/part2/StructuralAlignment.ipynb.