Harnessing Machine Learning for Predicting the Toxicity of Novel Antimicrobial Drugs

Many antimicrobial compounds effectively kill bacteria and fungi, but assessing their toxicity to humans is essential. Traditionally, this requires labour-intensive and ethically contentious tests on human cells and animals. This project will use machine learning to predict toxicity of drugs, reducing cost, time and animal use while accelerating drug development.

Award

MRes | MPhil | PhD | MSc by Research

Start Date

Usually February and October - at individual School's discretion

Research Opportunities

Summary

This project aims to create a machine learning model that can correctly predict the toxicity of novel antimicrobial compounds.

There are existing several data sets that can be utilised as the starting point to train such a model for such a purpose, including data from PubChem, ChEMBL and ToxCast. These provide the molecular structure of the compounds. A machine learning model trained on this data can then predict the toxicity of compounds it hasn’t seen before, utilising the fact that compounds with similar structures often have similar toxicity.

Various existing machine learning approaches to predicting the toxicity of compounds exist. This project will focus on developing new methods that are more effective at this task. In particular, one current limitation of the most recent deep learning approaches to predicting toxicity is that they require very large amounts of data to learn from accurately. One line of research during the project will be to tackle this bottleneck by adapting techniques which can reduce the amount of data needed, such as synthetic data generation, data augmentation, transfer learning, few-shot learning, or active learning.

There is some flexibility in the project, and a candidate from either a biosciences or computing background would be considered. The ideal student would have experience and interest in either machine learning or bioinformatics, or a related field such as data science. Strong computer programming skills would be very advantageous, as the project will involve the creation of novel methods, not just using existing tools. Knowledge relating to antimicrobial compounds or toxicity may be useful but is not required. The project is unlikely to involve significant lab work, and will mainly be computer based.

The supervisory team have the necessary skills and experience to oversee, guide and train the student in all aspects of the project.

Dr Bartlett works in the field of artificial intelligence and will oversee the machine learning aspects of the project. His role will be to provide advice on machine learning techniques to investigate and the adaptation of these to the problem domain.

Dr Welgamage is an experienced researcher in the field of microbiology. He will provide guidance on the antimicrobial problem domain – advising on the choice of data to use, designing lab-based assays to test predictions if necessary, and providing insight into real-world drug development needs.