pepdist

Installation

The Package can easily installed via pip. Just enter:

pip install pepdist

and the package and all dependencies should be installed.

To use the package just import it with:

import pepdist
from pepdist import distance, similarity

Getting started

The package consists of two major sub-packages:

• Similarity : This package uses a substitution- or other scoring matrices two obtain a measurement of peptide similarity. The nearest neighbor of a given query peptide in a big set of peptides can then be computed by a fast trie-based branch and bound algorithm.
• Distance : This package uses a feature vector representation of the peptides given for example by the AAindex database. The distance between peptides can then be defined as the euclidean distance between these vectors. A locally sensitive hashing can then obtain the approximated nearest neighbor.

While the code was designed for peptide comparison, it can also be used for other purposes.

Similarity

The Trie can be filled with words of different length, but the nearest neighbor search only compare sequences with the same length. Additionaly the positions can be weighted by a real value.

>>>from pepdist import similarity

>>>x = ["AAAGG", "AGA", "WWYYA", "WWWW", "YYY", "AYA"]
>>>trie = similarity.Trie(x)

>>>trie.k_nearest_neighbour("AAA", similarity.blosum62)
[('AGA', 0.6172133998483676)]
>>>trie.k_nearest_neighbour("WWYYY", similarity.blosum62)
[('WWYYA', 0.8198127976897006)]
>>>trie.k_nearest_neighbour("WWYYY", similarity.blosum62, weights=[1,1,1,1,0])
[('WWYYA', 1.0)]

The kmerTrie can be used in a similar way. However given a number k, it also saves all k-mers for all word in the trie. This can be used to not only score equal length peptides, rather you can find the best scoring substring of length k.

Distance

The Aaindex class parse the AAindex database into python dictionaries. This indices map a amino acid to a real value and therefore can be used to map petides to a feature vector representing a physical or chemical property. The distance between two peptides can then be represented as the euclidean distance of the feature vectors. However encoding a peptide with multiple indices can lead to very high dimensional feature vectors. For an efficent approximate nearest neighbour search a locally sensitive hashing was implementet which allows blazing fast search times.

Here are some example of how to use it:

>>> from pepdist import distance
>>> x = ["AAAGG", "AGAAA", "WWYYA", "WWWWA", "YYYWW", "AYAWW"]
>>> AA = distance.Aaindex()
>>> desc = distance.IndexDescriptor(x, [AA["GEIM800103"]])
>>> desc.
>>> desc.sequences
['AAAGG', 'AGAAA', 'WWYYA', 'WWWWA', 'YYYWW', 'AYAWW']
>>> desc.descriptors
[array([1.55, 1.55, 1.55, 0.59, 0.59]), array([1.55, 0.59, 1.55, 1.55, 1.55]), array([1.86, 1.86, 1.08, 1.08, 1.55]), array([1.86, 1.86, 1.86, 1.86, 1.55]), array([1.08, 1.08, 1.08, 1.86, 1.86]), array([1.55, 1.08, 1.55, 1.86, 1.86])]
>>> desc.change_norm_method(distance.z_normalize)
>>> desc.descriptors
[array([ 1.02132811,  1.02132811,  1.02132811, -1.01708142, -1.01708142]), array([ 1.02132811, -1.01708142,  1.02132811,  1.02132811,  1.02132811]), array([1.67956452, 1.67956452, 0.02335678, 0.02335678, 1.02132811]), array([1.67956452, 1.67956452, 1.67956452, 1.67956452, 1.02132811]), array([0.02335678, 0.02335678, 0.02335678, 1.67956452, 1.67956452]), array([1.02132811, 0.02335678, 1.02132811, 1.67956452, 1.67956452])]
>>> lsh = distance.LSH(5)
>>> lsh.add(desc.sequences, desc.descriptors)
>>> lsh.nearest_neighbour(desc.descriptors[0])
('AAAGG', 0.0)

Documentation

A full read the docs documentation can be accesed by: https://manuelgloeckler.github.io/pepdist/