Representation learning for anti-microbial resistance

Overview

This project is about using neural network models help us understand Anti-Microbial Resistance (AMR), a phenomenon in which bacteria adapt to reduce the effectiveness of antibiotics, usually through a process known as Lateral or Horizontal Gene Transfer - where genes are included in the organism from other sources.

Our focus will be on learning compact vector representations of biological sequences known to be associated with AMR genes. By encoding DNA sequences in this way we can more rapidly identify AMR genes in new strains, and perform more extensive surveys of existing sequences, enabling better understanding of the mechanisms involved.

The treatment of infectious diseases was revolutionised in the twentieth century through the invention of antibiotics such as penicillin and its derivatives, enabling effective action against potentially life-threatening bacterial pathogens.

Over time, the effectiveness of some of these drugs has been compromised by over-prescription and the rise of drug resistant strains such as Methicillin Resistant Staphylococcus aureus (MRSA ), commonly known as Golden Staph. Hospital acquired infections of this type are a significant global health problem and are potentially very dangerous, causing serious illness and even death to patients unfortunate enough to be affected.

It is therefore important to understand better how drug-resistant strains are created, and to develop tools to rapidly identify new threats as they appear.

Research activities

Work on this project will involve the application of representation learning models to curated sets of biological sequences. The work builds upon a short student project based on applying Word2Vec in this domain, and will potentially apply and extend novel representation learning methods developed as part of a recent PhD thesis.

The activities will depend on the level of the student, but minimally there will opportunities to:

• train and analyse machine learning models for AMR gene detection
• extend the approach to alternative representation learning methods
• apply the models to large scale data sets and investigate patterns of AMR evolution.

Outcomes

The outcomes will depend on the level of the project, but over time we want to create new methods for the detection and analysis of AMR genes and to greatly speed up the computations involved.

Ideally we will also conduct a large scale survey of known sequences for perhaps previously unrecognised AMR content, and to better understand the evolution of a class of AMR strains.

Skills and experience

You should have:

• good programming skills
• an ability to work with complex datasets and to understand machine learning algorithms
• a willingness to learn the biology needed to understand the domain.

Most of our students have studied or are studying computer science, but we welcome anyone who comes with a mix of skills that can attack the problem. Those with a joint degree involving molecular biology and computer science are especially welcome.

It isn't necessary for you to be an extraordinary software developer but you need to comfortable in Python or C# or Java or F# or other modern languages. This isn't a project where you can learn to program. We will teach you the biology and the machine learning as the project takes shape.

Please get in touch if this sounds like you.

Note: You may be able to apply for a scholarship if undertaking this project for a full-time Honours or PhD programme.

Scholarships

You may be eligible to apply for a research scholarship.

Explore our research scholarships

Keywords

• neural networks
• anti-microbial resistance
• bioinformatics
• machine learning

Contact

Please contact the supervisor for more information.