Back to: Study
AI-based exploration of crystal spaces to accelerate drug discovery
Student: Jonathan Balasingham
Supervisory Team: Dr Viktor Zamaraev (Computer Science), Dr Vitaliy Kurlin (Materials Innovation Factory), Prof Andy Cooper (Chemistry)
The crystal structure of a given pharmaceutical greatly impacts its physical properties. Aspects like solubility, stability, and dissolution rate are directly tied to the drug's structure and obtaining optimal characteristics is crucial for effective development. Though there exist well-established techniques for this goal, they are often computationally expensive and result in far too many possible predictions. As such, we wish to offer an alternative method of determining crystalline structure and properties. Newly designed crystal invariants allow us to unambiguously describe a crystal's structure and provide a medium to represent its topology. Using these invariants, we wish to engineer features that can be incorporated into a learning algorithm that can accurately determine the various properties of a particular form of a crystalline drug. Furthermore, we wish to be able to invert this process and provide a model capable of producing candidate crystal structures that correspond to a set of desired properties, while avoiding over-prediction. By using a machine learning approach we hope to reduce the computational cost of this process and provide a system capable of being applied to large sets of crystals.
Understanding bacterial resistance by machine learning from genetic data
Student: Alessandro Gerada
Supervisory Team: Prof. William Hope (Department of Pharmacology and Therapeutics), Dr Vitaliy Kurlin (Department of Computer Science), Prof. Steve Paterson (Department of UoL Genetics)
Bacterial resistance to antibiotics remains one of the biggest challenges in medicine. Whole genome sequencing continues to become more accessible, and produces a complete reading of a bacterium’s genetic code. This aim of this project is to develop mathematically justified algorithmic predictions of bacterial susceptibility to antibiotics using only the bacterial genome. We aim to collect 800 clinical bacterial strains through our local NHS partner - Liverpool Clinical Laboratories. These strains will be sequenced in collaboration with the Centre for Genomic Research using state-of-the-art Next Generation Sequencing techniques. The data produced will be used as an input for a machine learning model that predicts susceptibility of bacteria to a panel of antibiotics. The output will be a gold standard measure of antibiotic concentration required to suppress bacterial growth and therefore inform on the clinical suitability of the antibiotic for treating a patient’s infection. One of the challenges will be the high dimensions of data produced from genome sequencing. The machine learning approach will showcase the use of whole genome sequencing data in the understanding and management of antimicrobial resistance.
A-Eye: Integrating deep learning and in-silico modelling to optimize diagnosis and treatment of wet age-related macular degeneration
Student: Remi Hernandez
Supervisory Team: Dr Wahbi El-Bouri (Department of Cardiovascular and Metabolic Medicine), Professor Yalin Zheng (Department of Eye and Vision)
Wet age-related macular degeneration is a blinding disease caused by the growth of blood vessels in the tissue lining the back of the eye, the retina. These immature vessels not only grow in areas normally avascular but are also often leaking blood and fluids in the retina. Combined, those two aspects provoke deformations of the tissue that eventually causing the death of cells essential to vision. This causes quick and irreversible loss of sight. The treatment for this disease consists in regular injections meant to inhibit the growth of new vessels. However, an early diagnosis is essential to maintain vision, and even still the response to treatment is varied. In part, this is due to the abnormal vasculature growing in different layers of the retina. We hope to help the interpretation of ophthalmic images by applying machine learning algorithm to classify the disease according to the location of the abnormal vasculature. Additionally, we will try to improve on the treatment by using mathematical models of oxygen and drug availability within the retina. More precisely, we want to be able to run simulations of the treatment on computer for a given type of patient and use those simulations to improve on the current treatment strategies.
Diagnosis of Baby Neurodevelopmental Disorders Using Millimetre-waves
Student: Anthony Nzegbuna
Supervisory Team: Dr Jiafeng Zhou (EEE), Professor Xiaowei Huang (Computer Science), Mark Turner (Institute of Life Course and Medical Sciences)
The use of several high frequency bands of radar for the extraction of some medical information (such as cardiorespiratory features) has in recent years been of huge interest to researchers. Millimetre wave radar is non-invasive, non-contact and non-ionising, therefore very suitable for applications in paediatric medicine. Very small vibrations of the heart and lung including movements of other body parts can be detected, monitored and classified with greater reliability using the millimetre-band radar signals. The primary focus of this research is the deployment of Machine Learning algorithms towards characterising radar signal outputs from these two vital human organs together with certain body gestures which are critical to the study and recognition of infantile neurological development. Through a combination of medical AI and radar technology, the project will also examine the prospect of identifying one or two neurological conditions as revealed by the human-induced actions: following successful characterisation of attendant radar signatures.
Toward Autonomous Cannulation in Endovascular Intervention
Student: Tudor Jianu
Supervisory Team: Dr Anh Nguyen (Computer Science), Dr Sebastiano Fichera (Engineering)
Many robotic devices for endovascular intervention have been developed recently. They typically have a decoupled architecture in which a master device is controlled by the surgeon and the information inputted is routed to a slave device that executes the action (e.g. translation/rotation). Various assistive functions, such as force estimates and haptic feedback, or real-time X-Ray (fluoroscopy), have been developed, some with the help of deep learning. However, those systems handle low-level autonomous aspects while the operation is still performed manually by the surgeon. The goal of this PhD research is to create solutions for autonomous cannulation and tracking based on machine learning, bioelectrical localisation, and robotics. The cutting-edge autonomous cannulation and tracking technology will be integrated into a cutting-edge robotic steering system for cooperative catheter placement. The technology will have numerous degrees of freedom for precise control of the insertion process, haptic feedback for a better user experience, and the option for human or autonomous robotic insertion. The ultimate product will be a new, self-contained, and radiation-free endovascular surgical system.
Developing deep graph neural networks for prediction of drug toxicity
Student: Nandini Gadhia
Supervisory Team: Dr Anh Nguyen (Computer Science), Dr Vitaliy Kurlin (Materials Innovation Factory)
I will apply artificial intelligence techniques to understand the impact of existing and proposed drugs on cell and protein function. Particularly, I will investigate phenotypic effects using graph-based representations. Graph techniques lend well to computational biology as their embeddings can capture biological features and the relationships between them. These can then be analysed to give recommendations on drug toxicity in various circumstances.