Machine Learning for Protein Function

TL;DR

This paper develops machine learning-based feature engineering methods and tools like NeuroPID and ProFET for identifying and classifying proteins with similar functions but low sequence similarity, enabling large-scale discovery and analysis.

Contribution

Introduces novel feature engineering frameworks and tools for protein classification, improving identification of functionally related proteins with minimal sequence similarity.

Findings

ProFET achieves state-of-the-art performance on benchmark datasets.

NeuroPID enables mass discovery of neuropeptides and NPPs.

Tools are applicable to diverse high-level protein functions.

Abstract

Systematic identification of protein function is a key problem in current biology. Most traditional methods fail to identify functionally equivalent proteins if they lack similar sequences, structural data or extensive manual annotations. In this thesis, I focused on feature engineering and machine learning methods for identifying diverse classes of proteins that share functional relatedness but little sequence or structural similarity, notably, Neuropeptide Precursors (NPPs). I aim to identify functional protein classes solely using unannotated protein primary sequences from any organism. This thesis focuses on feature representations of whole protein sequences, sequence derived engineered features, their extraction, frameworks for their usage by machine learning (ML) models, and the application of ML models to biological tasks, focusing on high level protein functions. I implemented the ideas of feature engineering to develop a platform (called NeuroPID) that extracts meaningful features for classification of overlooked NPPs. The platform allows mass discovery of new NPs and NPPs. It was expanded as a webserver. I expanded our approach towards other challenging protein classes. This is implemented as a novel bioinformatics toolkit called ProFET (Protein Feature Engineering Toolkit). ProFET extracts hundreds of biophysical and sequence derived attributes, allowing the application of machine learning methods to proteins. ProFET was applied on many protein benchmark datasets with state of the art performance. The success of ProFET applies to a wide range of high-level functions such as metagenomic analysis, subcellular localization, structure and unique functional properties (e.g. thermophiles, nucleic acid binding). These methods and frameworks represent a valuable resource for using ML and data science methods on proteins.