Clustering FASTA datasets using dimensionality reduction algorithms.

• BLAST: Convert FASTA file to protein database and search using FASTA file as query
• jackhmmer: PSI-BLAST-like iterative search of a FASTA query against itself

Results are stored in a abstracted pairwise, symmetric, similarity matrix.

Converts high dimensional similarity matrix into low-dimensional embedding while maintaining local manifold structure.

Embedding is stored as a two-dimensional numpy array.

Uses matplotlib to plot embedding with basic annotation tools.

Ensure that the file structure for the src/ folder is as follows: (It should already be correct!)

• src
  • pyclust.py
  • lib/
    • annotation.py
    • grouper.py
    • initrun.py
    • jsonconv.py
    • mds_calc.py
    • plotter.py
    • readargs.py
    • results_parser.py
    • tsne/
    • tsne_calc.py

• We recommend you set up a virtual environment and run the command pip install -r requirements.txt

• Then run pip install tsne. (The setup for tsne requires the numpy module)

• Alternatively, install all the packages in requirements.txt manually.

Requirements.txt contents:

Cython==0.22
matplotlib==1.4.3
scikit-learn==0.16.1
scipy==0.15.1
tsne==0.1.1

• Also, you must have a working installation of either BLAST or jackhmmer for all vs. all sequence alignment.

Create a folder for output storage. The name of the folder must be the name of the the FASTA file. FASTA file name must be in the format example.fas

• mkdir proteins for FASTA file proteins.fas

Move FASTA file to this folder and create a temp/ directory.

• mv proteins.fas proteins/
• mkdir temp/

Run all vs. all sequence comparison. BLAST output must be in table format under the name results.out. Jackhmmer output must be in table format under the name tbl.hits.

• jackhmmer --tblout tbl.hits proteins.fas proteins.fas

Navigate run pyclust.py from src/ directory with the following flags:

• -dir proteins name of data folder (e.g. proteins)
• -parse parse into similarity matrix
• -cluster cluster using SVD/t-SNE
• -plot plot using Matplotlib
• E.g. python /Path/to/src/pyclust.py -dir proteins -parse -cluster -plot