Machine learning Banner

Current PhD Student Projects

Current PhD Research

In the Visual Computing and Signal Processing group we welcome new PhD students and if you would like to discuss PhD research projects then you are welcome to contact the members of this group.

In the group we collaborate with industry and research institutes on a wide range of projects from Molecular Docking to medical applications. You can see a selection of our current PhD student projects listed here:

  1. GPU Accelerated molecular graphics techniques for Interactive Molecular Docking

  2. Cartoon rendering for real time, interactive docking

  3. Conformational clustering for receptor flexibility in Dockit

  4. Visual geo-location for agricultural robots

  5. Cardiac Vessel Tree Reconstruction from Biplanar Xray Angiograms

GPU Accelerated molecular graphics techniques for Interactive Molecular Docking

Participants

Prof. Stephen Laycock(CMP)

Prof. Steven Hayward (CMP)

Dr. Georgios Iakovou

Mr. Alexander Faulkner

Summary

Molecular Docking simulates the interaction and binding of two or more molecules to form one larger molecule. There is usually a larger molecule (the protein or receptor), and a smaller molecule (the ligand). Interactive Molecular Docking allows a user to control the molecules and explore how they interact during the binding process. It has applications in structure based drug design in the pharmaceutical industry as well as in education, and is intended to improve understanding of how and why molecules bind together. A key part of molecule interactions is flexibility and deformation of the molecules resulting from the forces they exert on each other. The simulation of the interactions and the generation of visuals are computationally demanding tasks, and are run on a GPU to gain the performance benefits of parallel processing. My work aims to improve existing implementations in the DockIT program to handle interactive flexible docking between larger molecules, allowing the user to perform fully-flexible protein-protein docking.

Molecular graphics

Funding

The Edward and Ivy Rose Hood Memorial Scholarship

Cartoon rendering for real time, interactive docking

Participants

Prof. Stephen Laycock (CMP)

Ms. Katerina Holdsworth

Summary

The cartoon representation shows the secondary structure of a protein. It is an abstract depiction which shows the protein backbone and is good for orientation. Interactive cartoon rendering involves finding the backbone hydrogen bonds, assigning the secondary structure and constructing the surface every frame. This process is run in parallel using the GPU, allowing for real time frame rates. The end goal of this research is to construct and render the cartoon representation for use, in conjunction with other molecular representations, within an interactive docking software (DockIt).

110K

Funding

UEA

Conformational clustering for receptor flexibility in Dockit

Participants

Prof. Stephen Laycock (CMP)

Prof. Steven Hayward (CMP)

Dr. Georgios Iakovou

Mr. Chao Liu

Summary

This project enhances molecular structure analysis by applying Gaussian Mixture Models (GMM) to identify dominant conformational states from molecular dynamics data. We project the structural fluctuations into a low-dimensional PCA space and cluster the resulting trajectories to capture key structural variations. To evaluate clustering quality, we introduce a compactness and distinctness-based scoring method using the average deviations of clusters and responsibility of each frame of structure belonging to each cluster.

The resulting clusters provide a diverse set of representative receptor conformations that are particularly valuable for flexible docking. These representative structures are directly integrated into Dockit, a molecular docking software designed to explore multiple binding modes through ensemble docking. By incorporating flexible receptor states identified through clustering, Dockit becomes more effective in modelling realistic receptor–ligand interactions, which improves docking accuracy and relevance for drug discovery.

This approach bridges computational and structural biology, offering industrial and academic collaborators a practical framework for analysing molecular flexibility and enhancing docking simulations.

GMM clustering

Funding

CSC

Visual geo-location for agricultural robots

Participants

Dr. Yingliang Ma (CMP)

Mr. Calvin John

Summary

The agri-robot industry relies heavily on GNSS  systems to localise (find the exact position of) robots in rural environments. Localisation is crucial in robot navigation, the more accurate the localisation the more precise the navigation of the robot will be. Localisation precision is particularly important in farming as there are tasks on farms that require cm levels of precision, such as precision spraying(for applying pesticides or nutrients to crops), harvesting, weeding etc. In accurate localisation systems make such tasks much more difficult and can potentially cause losses in harvest yield and thus a decrease in profit margins. To obtain high precision localisation, farmers will usually use GNSS RTK which are effectively highly accurate GPS readings which provide GPS coordinates that contain on average between 3 to 10 centimetres of error as opposed to the 1 to 3 metres of error seen in ordinary GNSS/GPS readings. The problem with GNSS RTK is that it’s typically quite expensive for startups, usually costing upwards of £1500 for industry grade receivers, this can commonly reach up to £15000 or more.

In this project we to create a visual geo-localisation system that is robust to signal interference and is also highly accurate. Whilst it is understood that purely visual methods will not be able to match GNSS RTK systems in terms of raw precision, we aim to get as close as possible to this level of precision.

visual geo-location diagram

Partners

uea cmp
University of Lincoln

Funding

UKRI A Spatio-chromatic model

Cardiac Vessel Tree Reconstruction from Biplanar Xray Angiograms

Participants

Dr. Yinglinag Ma (CMP)

Dr. Lin Xi (CMP)

Mr. Ethan Koland

Summary

Through vessel branching point detection and matching, we hypothesize that we can reconstruct high accuracy 3D representations, point clouds, of cardiac vessel trees. Project aims to develop and implant AI networks capable of replacing manual interventions. These models will allow real time, 15-20 FPS, registration and reconstruction of Vessel trees within 5 mm accuracy to increased dimensionality and information during surgical operations increasing patient outcomes.

Current research involves manual intervention for point detection and matching with deep learning based reconstruction techniques. Current approaches toward reconstruction involves several key draw backs. Rather than use matched biplanner images for reconstruction we aim to develop a method that works regardless of motion synced images.

We also hypothesis that the use of vessel branching point also provides key points for registration of 3D representations on x-ray video streams providing a real time. This would allow online reconstruction of the vessel tree with real time registration regardless of cardiac or respiratory motion.

Funding

EPSRC logo

Publications

[1] Yao, Linlin et al. (Nov. 2023). “TaG-Net: Topology-Aware Graph Network for Centerline-Based Vessel Labeling”. In: IEEE Transactions on Medical Imaging 42.11, pp. 3155–3166. ISSN: 1558-254X. DOI: 10.1109/ TMI.2023.3240825. URL: https:// ieeexplore.ieee.org/document/10032183

[2] Wang, Yiying, Abhirup Banerjee, Robin P. Choudhury, et al. (Dec. 13, 2024). DeepCA:Deep Learning-based 3D Coronary Artery Tree Reconstruction from Two 2D Non-simultaneous X-ray Angiography Projections. URL: http://arxiv.org/abs/2407.14616

[3] Banerjee, Abhirup et al. (Apr. 2020). “Point-Cloud Method for Automated 3D Coronary Tree Reconstruction From Multiple Non-Simultaneous Angiographic Projections”. In: IEEE Transactions on Medical Imaging 39.4. Conference Name: IEEE Transactions on Medical Imaging, pp. 1278–1290. ISSN: 1558-254X. DOI: 10.1109/ TMI. 2019.2944092. URL: https://ieeexplore.ieee.org/document/8864089

Current PhD Student Projects