proteinko

Encode protein sequence as a distribution of its physicochemical properties.

• Introduction
• Methods
• Installation
• Usage
  • Example 1
  • Example 2
• Release Notes
  • release 5.0

Introduction

Protein is as sequence of amino acid residues connected by peptide bonds. Each amino acid residue is characterized by a unique combination of its physical and chemical properties. proteinko takes advantage of this to represent protein sequence as a spatial distribution of the physicochemical properties of its amino acid residues, capturing the complementing or cancelling effect of neighbouring amino acid residues.

proteinko enables numerical representation of a protein sequence while preserving the information of its underlying physicochemical properties. This allows the investigation of relationships and interactions between proteins as well as potential discovery of underlying physicochemical properties which facilitate those interactions.

Methods

proteinko implements a fairly simple algorithm. The protein sequence is mapped to a vector V representing a distribution of a certain physicochemical property of the entire protein. Each amino acid residue Ai is modeled independently as a Gaussian curve Gi and scaled by the corresponding value from the encoding scheme. Gi is mapped to the slice of V which is centered at a position correspondig to the position of Ai in the sequence and which spans L neighbouring slices on each side. The overlap allows to sum the complementing or cancelling effects that the neighbouring amino acid residues may exert on the local physicochemical property of the protein. The extent of overlap is determined by two factors: overlap distance (L) and sigma factor. Overlap distance determines how many neighbouring slices Gi spans on each side. Sigma determines the shape of the Gaussian curve of each of the amino acid residues (see example). Both of these parameters proteinko accepts as function arguments allowing users to modify the shape of final distribution as needed.

Instalation

pip install proteinko

Usage

proteinko implements two functions: model_distribution and encode_sequence. Both functions have encoding_scheme parameter which accepts a python dictionary with amino acid one-letter codes as keys.

Example 1:

from proteinko import model_distribution, encode_sequence
import matplotlib.pyplot as plt
from pyaaisc import Aaindex


sequence = 'MEEPQSDPSVEPPLSQETFSDLWKLLPENNVLSPLPSQAMDDLMLSPDDIEQWFTEDPGP'
encoding_scheme = Aaindex().get('ARGP820101', dbkey='aaindex1').index_data

dist_1 = model_distribution(sequence, encoding_scheme, overlap_distance=2, sigma=0.4)
dist_2 = model_distribution(sequence, encoding_scheme, overlap_distance=3, sigma=0.8)
encoded = encode_sequence(sequence, encoding_scheme)

fig, ax = plt.subplots(3, 1, sharey=True, figsize=(12,5))
ax[0].plot(dist_1)
ax[0].grid()
ax[0].set_xticklabels([])
ax[0].set_title('Modeled distribution, overlap_distance=2, sigma=0.4')
ax[1].plot(dist_2)
ax[1].grid()
ax[1].set_xticklabels([])
ax[1].set_title('Modeled distribution, overlap_distance=3, sigma=0.8')
ax[1].set_ylabel('Hydrophobicity index - ARGP820101')
ax[2].bar(range(len(encoded)), encoded)
ax[2].grid()
ax[2].set_xticks(range(len(sequence)))
ax[2].set_xticklabels([x for x in sequence])
ax[2].set_title('Sequence')

plt.show()

Example 2

from proteinko import model_distribution
import matplotlib.pyplot as plt
from pyaaisc import Aaindex


sequence = 'MEEPQSDPSVE'
encoding_scheme = Aaindex().get('ARGP820101', dbkey='aaindex1').index_data

dist = model_distribution(sequence, encoding_scheme, overlap_distance=2, sigma=0.4)
sampled_dist = model_distribution(sequence, encoding_scheme, overlap_distance=2, sigma=0.4, sampling_points=16)

fig, ax = plt.subplots(2, 1, figsize=(6,4))
ax[0].plot(dist)
ax[0].grid()
ax[0].set_xticklabels([])
ax[0].set_title('Modeled distribution')
ax[0].set_ylabel('Hydrophobicity index')

ax[1].bar(range(16), sampled_dist)
ax[1].grid()
ax[1].set_xticklabels([])
ax[1].set_title('Sampled distribution')
ax[1].set_ylabel('Hydrophobicity index')

plt.show()

Release Notes

release 5.0

Algorithm changes:

• Number of overlaping neigbouring amino acid residues has been added as function argument and default value set to overlap_distance=2.
• Default sigma value has been changed from 0.8 to 0.4.
• Normalization and standardization of modeled distribution are deprecated. No pre or post processing is applied.
• Scaling factor has been decreased from 100 to 40, reducing the number of computations and increasing the performance of algorithm.

Major code changes:

• Proteinko class has been removed and algorithm is implemented under model_distribution function.
• New function encode_sequence has been introduced which simply encodes sequence with values provided in the encoding table.
• Encoding tables are now passed as python dictionaries instead of pandas dataframe.
• Use of pandas and scipy packages has been replaced with python functions making the code more lightweight and increasing the performance of algorithm.

Minor code changes:

• vlen parameter has been renamed to sampling_points because it is the number of points to sample from final distribution.
• schema parameter has been renamed to encoding_scheme.