peptidy 0.0.1

A Python library that converts peptide sequences to matrices for machine learning.

peptidy

Welcome to peptidy — a tiny and tidy python library to vectorize peptide sequences (and proteins) for machine learning applications!

peptidy bridges the gap between peptide sequences and machine learning applications by converting peptide sequences into numerical vectors with a single function call. peptidy is obsessed with the simplicity and tidiness — it is designed to be as accessible as possible for everyone.

Key features of peptidy include:

• Simple: The programming interface of peptidy is designed to be as simple as possible and well-documented to minimize entry barriers. Converting a peptide sequence into a numerical vector is just a single function call.
• Tiny: peptidy is written in pure Python (>=3.6) and free from external dependencies. It is designed to be light-weight.
• Broad: peptidy supports a wide range of encoding schemes for both discriminative and generative applications, e.g., one-hot encoding, BLOSUM62, and physicochemical descriptors. Moreover, peptidy can handle peptides with post-translational modifications.

Installation :hammer_and_wrench:

pip install peptidy

Simple :shrug:

Quick Start :rocket:

The most important module in peptidy is peptidy.encoding. This module provides functions to convert peptide sequences into numerical vectors. Five encoding methods are implemented: aminoacid_descriptor_encoding, blosum62_encoding, label_encoding, one_hot_encoding and peptide_descriptor_encoding. All of those functions take a peptide sequence as input and return a numerical vector or matrix. An extensive documentation is available for all encoding methods :book:

Here are some quick examples to start with:

# Represent a peptide sequence with label encoding
import peptidy
peptide = "ACD"
padded_labels = peptidy.encoding.label_encoding(peptide, padding_len=5)  # [1, 2, 3, 0, 0]

There you go! The peptide "ACD" is converted into a list of labels, padded, and is ready to be fed into a deep learning model.

Post-translational Modifications :sparkles:

peptidy can handle peptides with post-translational modifications, such as phosphorylation, acetylation, and glycosylation. You can find a full list of supported modifications in the documentation.

# Represent a peptide sequence with one-hot encoding and phosphorylation
peptide = "ACS_pD"  # S_p stands for phosphorylated Serine
padded_one_hot = peptidy.encoding.one_hot_encoding(peptide, padding_len=5)  # (5, 29)

Computing Peptide Descriptors :desktop_computer:

Peptides are commonly represented by their physicochemical properties in machine learning models and peptidy provides functions to facilitate those models. In addition to providing peptidy.encoding.peptide_descriptor_encoding function, peptidy also exposes a module, peptidy.descriptors with many peptide descriptors, e.g., charge, hydrophobicity, and instability index. The full list and descriptions of supported descriptors is available on the documentation page.

# Compute molecular weight of a peptide
peptide = "ACS_pD"
descriptors = peptidy.descriptors.molecular_weight(peptide)  # 474.375

Concluding Remarks

This was a quick introduction to capabilities of peptidy. But peptidy can empower much more! We provide classification and generation examples using xgboost, tensorflow, and keras, where peptidy smooths the computations. Feel free to check our documentation for more information and examples and create issues or pull requests on our GitHub repository for added functionalities or bug reports.

See you in the forums! :wave: