Gnocis: An integrated system for interactive and reproducible analysis and modelling of cis-regulatory elements in...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression regulation relies on cis-regulatory elements (CREs) like promoters and enhancers.
• Predicting CREs computationally often involves machine learning and DNA sequence features.
• Existing Python tools lack integration for combining diverse feature sets with machine learning for genome-wide CRE prediction.

Purpose of the Study:

• To introduce Gnocis, a Python package designed to streamline the analysis and modeling of CRE sequences.
• To facilitate the integration of various DNA sequence feature sets with machine learning models for genome-wide CRE prediction.
• To provide a versatile tool for DNA bioinformatics tasks, including sequence handling and preparation.

Main Methods:

• Gnocis implements extensible APIs for combining feature sets and machine learning models.
• It includes base feature sets like motif pair occurrence frequencies and k-spectrum mismatch kernel.
• Integration with Scikit-learn and TensorFlow supports state-of-the-art machine learning, including a novel Deep-MOCCA neural network architecture.

Main Results:

• Gnocis successfully integrates diverse feature sets and machine learning algorithms for CRE prediction.
• The package was used to model D. melanogaster Polycomb/Trithorax Response Elements (PREs) using Convolutional Neural Networks (CNNs).
• Deep-MOCCA demonstrated moderate to high generalization without prior motif knowledge.

Conclusions:

• Gnocis provides a unified platform for genome-wide CRE prediction, simplifying complex analyses.
• The package is adaptable for new CRE modeling tasks and across different organisms.
• Gnocis, implemented in Cython for performance, is available via pip and its source code is on GitHub.