Projects available for 2027-28 entry
MRes students work on their research project throughout the year. You can apply for one of the projects listed below, or contact the Course Director, Dr Jun Ishihara (j.ishihara@imperial.ac.uk) to discuss a different project.
Applications will be considered in three rounds. We encourage you to apply in Round 2 or 3 for the best chance to be considered for your preferred project.
Visit the course page for full instructions and deadlines.
List of Projects
Projects available for 2027-28 entry
- Professor Jimmy Moore
- Dr Kaushik Jayaram
- Dr Laki Pantazis
- Dr Christopher Rowlands
- Dr Guang Yang
- Dr James Choi
- Dr Jun Ishihara
- Dr Pedro Ballester
- Professor Darryl Overby
- Professor Mengxing Tang
- Professor Rylie Green
Professor Jimmy Moore
Profile: https://profiles.imperial.ac.uk/james.moore.jr
Contact details: james.moore.jr@imperial.ac.uk
| Project title | Description |
| Lymph node physiological mass transport | Many important immune processes occur in lymph nodes, but we actually know little about their structure. This is important because it determines patterns of mass transport which assure that antigens are presented to the appropriate cells. Based on a combination of high-resolution imaging protocols that quantify the structure of human lymph nodes, we will construct models of mass transport of chemokines and antigens in collagen bundle conduits, parenchyma and blood vessels. |
| Lymph Node Implant for Breast Cancer-Related Lymphoedema | A large percentage of breast cancer patients who undergo lymph node resection develop an incurable swelling of the arm called lymphoedema. We are developing an implant to replace the fluid delivery characteristics of lymph nodes. We have developed a lymph node implant that releases a growth factor to regenerate the damaged lymphatic vessels. We would like to visualize the release and flow of the growth factor once it is implanted in the tissue. We can do this using a microfluidic chip that simulates fluid flow like in tissue. The goal of this project is to use a fluorescence microscope to live image the microfluidic device as the fluorescent-labelled growth factor flows through it over time. This will help us understand where the growth factor travels to within the surgical site. |
| MRI phantom for testing lymphatic vessel imaging sequences | There are no means for measuring pressure, flow or diameter reliably in any lymphatic vessel without surgical exposure. While the vessel diameters are on the order of MRI spatial resolution, their unique flow dynamics profile offers opportunities to develop diffusion-based imaging sequences to distinguish them from blood vessels and interstitial fluid movement. We aim to develop a phantom that reproduces these flow patterns to aid in the development of better imaging sequences. |
Dr Kaushik Jayaram
Profile: https://profiles.imperial.ac.uk/k.jayaram
Contact details: k.jayaram@imperial.ac.uk
| Project title | Description |
| Modeling of Insect-scale Shape Morphing Robots | The project involves modeling the kinematics and dynamics of insect-scale bioinspired shape morphing robots in IssacGym/ Mujoco to create digital twins. These models will be used for training machine learning algorithms and for developing AI bioinspired controllers for navigating complex terrains. Will involve collaborations with Computer Science Department at Imperial and at ETH Zurich. Prior experience with physics-based modeling softwares in a must. |
| Digital twins of spiders | The project involves modeling the kinematics and dynamics of arthropods in Unity/ Mujoco to create digital twins using high-fidelity tracking data collected (DeeplabCut, Replicant) from spiders moving on a treadmill at varying inclinations (vertical, lateral and upside down). These models will inform the creation of new bioinspired gaits to be deployed on insect scale robots. Involve collaboration with other insect labs in department. Expect a strong background in data processing, programming, AI/ML. Prior experience with computer graphics and modeling is an advantage. |
Dr Laki Pantazis
Profile: https://profiles.imperial.ac.uk/p.pantazis
Contact details: p.pantazis@imperial.ac.uk
| Project title | Description |
| Beyond fluorescence: watching genetically-encoded bioharmonophores assemble in live mammalian cells | Fluorescent proteins have shaped modern cell biology, but they come with trade-offs: they bleach, saturate, and blur under high-intensity light. These limitations cap what we can see—especially when tracking fast or subtle dynamics over long periods. What if we could express a label that never fades, never saturates, and produces a clean, quantifiable signal—one that’s visible through even the densest cellular environments? This project gives you the opportunity to build and validate a new class of genetically encoded imaging probes—bioharmonophores—in live mammalian cells. These are not fluorescent proteins. Instead, they are small peptides that self-assemble into non-centrosymmetric nanocrystals inside protein shells, generating second-harmonic generation (SHG) signal: a narrow, photostable, and unbleachable optical readout. At the heart of the system is a synthetic expression circuit: peptides with strong SHG potential are targeted to encapsulin nanocompartments, where they remain inert until released by a genetically encoded TEV protease. This two-part design allows for controlled liberation and local concentration of SHG-active sequences, triggering their self-assembly into crystalline structures that produce strong SHG signal when imaged. This project focuses on the critical proof-of-principle: can we trigger SHG-active peptide self-assembly inside living human or mouse cells? This has never been shown. Success would confirm that bioharmonophores can be expressed, activated, and imaged in standard cell culture, opening the door to genetically programmable, multiplexed SHG imaging in deep tissues. Computational/Practical student experience You will: • quantify SHG signal strength across time and conditions using an existing Zeiss two-photon microscope platform • analyse optical signatures to correlate peptide identity, structure, and signal intensity. This project gives you hands-on experience with synthetic gene circuits, nonlinear optics, mammalian cell engineering, signal quantification, and peptide design—a complete pipeline from molecule to imaging outcome. Outcomes By the end of the project, you will produce the first demonstration that genetically encoded bioharmonophores can self-assemble and generate SHG signal in live mammalian cells. |
| Programming the first cancer cell: engineering spatially confined CRISPR to trigger and track oncogenic mutations in living organoids | Cancer begins when genetic change arises in individual cells embedded within otherwise normal tissue. Yet conventional genome-editing approaches perturb populations simultaneously, making it difficult to observe how one newly mutant cell interacts with its neighbours and reshapes a tissue. Our laboratory is developing an optically controlled CRISPR platform to overcome this limitation. Cas9 is sequestered through the photocleavable protein PhoCl and released by Primed Conversion, in which intersecting laser beams provide highly localised optical activation. Recent work in our lab established proof-of-principle light-dependent Cas9 activity and demonstrated spatially confined Primed Conversion within individual cells in 3D organoids. This project will engineer a next-generation light-gated CRISPR circuit with improved OFF-state repression and high light-induced dynamic range. The student will quantitatively characterise alternative circuit designs using molecular cloning, flow cytometry and live-cell microscopy, then transfer the optimised system into a 3D organoid model to test single-cell-restricted genome editing. Depending on progress and student interests, the project will extend to perturbing a disease-relevant endogenous gene, such as APC, and longitudinally tracking the resulting clone within surrounding wild-type tissue. The project combines synthetic gene-circuit engineering, CRISPR, optogenetics, organoid biology, advanced microscopy and quantitative image analysis. Principal Supervisor Name Periklis (Laki) Pantazis Principal Supervisor Email p.pantazis@imperial.ac.uk Secondary Supervisor Name Professor Mark Isalan Secondary Supervisor Email m.isalan@imperial.ac.uk |
| Towards single-cell precision ultrasound neuromodulation using optically traceable Bioharmonophores | Focused ultrasound provides a powerful non-invasive approach to neuromodulation, but its spatial resolution remains constrained by the acoustic focal volume, which typically encompasses many cells. This project will investigate a different strategy: using ultrasound as a broad, subthreshold energy source while defining the site of neuronal activation through spatially localised piezoelectric nanotransducers. We will test biodegradable peptide-based bioharmonophores as optically traceable nanotransducers. Bioharmonophores are polymer-encapsulated, self-assembling aromatic peptide nanoparticles that generate strong second-harmonic-generation (SHG) signals and biodegrade in biological environments.¹–³ Because ordered diphenylalanine assemblies exhibit pronounced piezoelectric properties,â´â€“ⶠwe hypothesise that suitably engineered bioharmonophores can convert ultrasound-induced mechanical stress into a highly localised electrical stimulus. The central hypothesis is that bioharmonophore-associated neurons can be activated at ultrasound exposures below those required for direct ultrasound neuromodulation, while neighbouring particle-negative neurons exposed to the same acoustic field remain inactive. If achieved, stimulation precision would be determined primarily by nanotransducer localisation rather than by the ultrasound focal volume. |
| Synchronised to Silent: Watching Piezo1 Signalling Break Down Under Shear in Real Time | Cells in epithelial tissues do more than just survive mechanical stress—they interpret it, synchronise their responses, and adapt. One of the most powerful molecular sensors orchestrating this process is Piezo1, a mechanically activated ion channel that converts shear stress into calcium signals. When activated, Piezo1 channels trigger transient calcium influxes, which can be monitored using live-cell biosensors. But how exactly do individual cells encode mechanical input? How do they coordinate their responses across space and time—and what happens when that coordination fades? This computationally based project gives you the opportunity to answer these questions using a rich, high-quality dataset of time-lapse fluorescence images acquired from GenEPi biosensor-expressing epithelial cells under shear stress. GenEPi is a genetically encoded calcium reporter that lights up when Piezo1 channels open, enabling direct, real-time readouts of mechanosensitive activity in live tissues. Crucially, the data have been collected under defined flow conditions using a custom-designed microfluidic platform, meaning you will be working with dynamic, structured datasets that closely reflect physiological mechanical environments. What makes this project particularly compelling is the discovery that Piezo1 activation is not only force-dependent, but also transient and coordinated across cells—until it isn’t. With continued or repeated stimulation, the system begins to lose its coherence: signal amplitudes decline, synchronisation breaks down, and the tissue shifts from a unified responder to a fragmented, desynchronised field. What governs this transition? Can it be predicted or quantified? These are not just questions of cell signalling—they touch on the principles of collective behaviour, adaptation, and signal fidelity in living systems. Your role in this project will be to explore, quantify, and model these dynamic transitions using image processing and computational analysis. By extracting time-resolved GenEPi fluorescence from single cells and tracking those responses across multiple stimulation cycles, you’ll uncover how shear stress modulates both the strength and structure of Piezo1 activity. You’ll also compare how different flow conditions—low vs. high shear—and cytoskeletal states affect signal propagation, adaptation, and decay. The work will involve building or adapting pipelines for image segmentation, signal extraction, time-series analysis, and population-level synchrony metrics. Computational/Practical student experience You will: - process and analyse large-scale time-lapse imaging datasets of GenEPi-expressing cells under controlled shear stress, - use Python, MATLAB or ImageJ/Fiji to extract ΔF/F₀ calcium signals from individual cells and clusters, - quantify signal dynamics over time: amplitude, rise time, duration, and frequency, - develop computational tools to measure intercellular synchronisation, using pairwise correlation, signal entropy, or phase-locking indices, - perform comparative analyses across shear conditions and treatment states to assess signal adaptation and coordination loss. This is a fully desk-based, computational project requiring no lab work or cell culture. You will work with already-acquired, publication-quality datasets and develop analytical methods that directly contribute to an active manuscript. The project offers deep training in image analysis, time-series data handling, and biological signal interpretation—skills with wide relevance in both academic and biotech contexts. Outcomes By the end of the project, you will have generated a series of analytical outputs—plots, heatmaps, and coordination metrics—that reveal how Piezo1 responses change over time and under stress. Your work will illuminate how force signals propagate through epithelial tissues, how collective behaviour is initiated and lost, and how cellular coordination is shaped by both mechanical inputs and intracellular architecture. These results will not only strengthen the current understanding of Piezo1 function but also provide a general framework for analysing other dynamic biosensor systems in mechanically active contexts. |
| Advancing Mechanobiology with ChemiGenEPi Biosensors | Cells are constantly pushed, stretched, and squeezed—and they feel it. Mechanobiology is the science of how cells sense and respond to forces, a process that drives development, organ function, and disease. At the centre of this force-sensing machinery is Piezo1, a pressure-sensitive ion channel. Our lab has already introduced GenEPi, the first genetically encoded Piezo1 activity reporter, but we want to go further. What if you could watch Piezo1 at work in real time, in living tissue, with sensors that are brighter, more stable, and tunable across colours? This project gives you the chance to build exactly that: ChemiGenEPi1.0, a brand-new chemigenetic biosensor. By combining the genetic precision of GenEPi with the dye-based flexibility of WHaloCaMP, ChemiGenEPi will let us see Piezo1 activity with unprecedented clarity—from single cells in culture to organs/organoids and whole animal models. Computational&Practical student experience You will: • test its performance in live-cell imaging assays (brightness, photostability, dynamic range), This project is hands-on at the interface of molecular engineering, synthetic biology, and cutting-edge imaging. You’ll gain experience in biosensor design, live-cell fluorescence and FLIM microscopy, and in vivo validation. Outcomes By the end, you’ll have helped create a next-generation tool for watching mechanosensitive signalling in action—one that could reshape how we study force in biology and even inspire future drug discovery. |
Dr Christopher Rowlands
Profile: https://profiles.imperial.ac.uk/c.rowlands
Contact details: c.rowlands@imperial.ac.uk
| Project title | Description |
| Diagnosing Disease with Speedy-Scanning Raman Readout | Ordinary microscopes can use light to diagnose diseases but they do so using a limited number of colours. If we replace the three colours of the spectrum with thousands of wavelengths in a spectrum, we can learn much more about each pixel in an image. This is the promise of Raman microscopy. Raman microscopy uses highly intense laser light to illuminate a sample. When the laser scatters from the sample, a very small fraction of it changes in wavelength, and this change is unique to a particular chemical bond. By mapping the sample using a spectrometer (which can image and quantify these wavelength changes) we can therefore gain chemical information about the sample, which is sufficient to diagnose a number of diseases, including (most notably) cancer. The downside of Raman microscopy is that it is slow, and while it can be sped up by increasing the illumination intensity, eventually this comes at the cost of damaging the sample. The key to increasing imaging speed is therefore to share the laser out over more pixels, recording from them in parallel. The student on this project will develop a new system that can illuminate dozens of points simultaneously, thus significantly speeding up the imaging process. This will involve working on an existing Raman microscope, modifying it to implement this new scanning process. The work will then progress to methods for rapidly diagnosing cancer using tissue biopsies. The ideal student on this project would have a reasonable background in programming, but all other skills and techniques can be taught. |
| Watching Sound - creating a new technique for stand-off ultrasound imaging | Ultrasound is one of the safest, cheapest and most powerful ways to image deep within the body. Compared to MRI it is fast, easy to use and significantly less onerous on the patient. Nevertheless, there are limitations which we are working to overcome. All current forms of ultrasound imaging require the user to place an ultrasound probe in contact with the skin. This in turn requires a skilled ultrasound technician to apply ultrasound gel and move the probe to image the organ of interest. A more elegant solution would be to use optical imaging to see the acoustic signal (as well as exciting it), thus removing the need for the technician, gel or even for the patient to lie on a bed. The acoustic signal could be simply recorded by imaging the patient's body with a very fast camera. The Rowlands lab is working on developing optical ultrasound detectors based on evanescent wave sensors; these are extremely sensitive to minute changes in the position of an array of nanoparticles, and thus to a passing acoustic wave. The student working on this project will help develop this new type of ultrasound detector, building the nanoparticle suspension, excitation optics and imaging / readout. The ideal student would have a background in the physical sciences or engineering, with a willingness to try new things and learn. The Rowlands lab is highly multidisciplinary, with lots of different researchers studying lots of different things, so new perspectives and approaches are encouraged. The student can be taught most (if not all) of the skills and techniques they will need to know. |
| Dynamic Dichroic Mirrors - making reprogrammable optical filters for stand-off chemical imaging | Hyperspectral imaging is used in applications from chemical weapon detection to cancer diagnosis, from fraud monitoring to industrial quality control. Currently wide-field camera-based hyperspectral imaging systems are based around single filters - you must know exactly what you're looking for in order to select the right filter. The Rowlands Lab is working on a new type of optical filter which can be reprogrammed at will, allowing arbitrary chemicals to be searched for, for example. Currently, the optical components have been assembled, but need to be tested and new materials tried out. The student on this project will be responsible for taking this system from prototype stage to working tool, and will have to develop a number of skills, from instrument development and debugging, to materials development and optimization and finally development of robust testing methods. There is also the potential for publication or intellectual property development, should groundbreaking advances be made. The ideal student on this project would have a willingness to learn, adaptability and some background in the physical sciences, engineering or computer science. That said, talented students from any background will be considered, and the relevant knowledge taught. |
| SIMaging the future | Optical microscopy is widely used in biomedicine as it is easy to use, safe for almost all samples, fast, and versatile. Unfortunately it has a physical limitation called the diffraction limit which prevents it from observing features smaller than ~100nm or so. Structured Illumination Microscopy (SIM) is a technique for increasing the resolution of an optical microscope. It works by taking nine images of a sample with different interference patterns applied to them, then reconstructing these images, using the Moire effect to work out what the underlying sample distribution looks like. The Rowlands lab has pioneered one of the fastest SIM systems in the world, and is in the process of developing version2 which should be practical an commercializable. The student on this project will be responsible for developing an improved version of the existing instrument that is more robust, compact and easier to use than the original, ultimately so that the instrument can be sold commercially. They should be comfortable handling expensive hardware with care, and ideally have a reasonable programming background (although all necessary skills on this project can be taught). The more important attributes are a willingness to learn and be self-motivated. |
| Turning tissue totally transparent | Within the last year there has been a breakthrough in our ability to control the transparency of live tissues. Typically, light scattering is the main reason why we can't see through things, and scattering is caused by microscale inhomogeneities in refractive index. By exploiting a mathematical relationship called the Kramers Kronig relations, we can now tune the refractive index of an object until it matches that of the rest of the local environment, making it essentially invisible. What the vast majority of people have failed to notice is that rather than just flooding a tissue with another medium to change its bulk refractive index, photopatterning can change the refractive index locally, allowing us to *remove* the refractive index differences, rather than just minimising them. The student will use the multiphoton microscope in the Rowlands lab to selectively photobleach a dye that has been infused in to a tissue, in an attempt to demonstrate that the refractive index can be changed to whatever value the user desires. This will involve wetlab work, potentially some programming and hardware design, and potentially even work with live tissues if the project is that successful. All of these skills can be taught, so the main requirement is a willingness to learn. |
| Edible holography | In the news recently it was announced that a common food dye, tartrazine, could be used to change the refractive index of an object. This is important because being able to structure refractive index on the microscale makes it possible to create true volume holograms, which can be observed in all directions and which are visually indistinguishable from the real object. The student on this project will be using tartrazine infused into Agar gels to create these volume holograms. By photobleaching the tartrazine in a carefully defined pattern, the refractive index could be changed to whatever the user desires, allowing these holograms to be built up one pixel at a time. And because the gels are made out of agar and tartrazine, they are (technically) edible. More seriously, these holograms can act as sensors, probes, vision correctors and so on. On the day-to-day basis the student will be creating gels, testing the photobleaching, designing holograms, and ultimately creating arbitrary three-dimensional holographic objects. This will combine optics, programming, web lab work, and analysis, but as all students in the Rowlands Lab are encouraged to make the project their own, the exact balance of these skills is flexible, and all can be taught. |
| Real laboratory automation | Laboratory automation usually refers to the use of robots to perform experiments and other repetitive tasks without user interaction, but another meaning could be to literally automate a laboratory; to control the lighting, shutters, fans and other noise sources automatically in response to voice commands, interlock conditions, number of users, lab temperature or any other condition. Until recently this will be an extreme expensive endeavour, but the proliferation of the low cost home automation equipment and open source projects such as home assistant mean that this can be explored at comparatively low cost. The goal of this project is therefore to make work in the laboratory safer, easier, more repeatable, and with better logging of conditions than previously has been possible. This will involve significant amounts of programming, a modest amount of hardware design, and possibly some interior decorations skills. Fortunately all of these can be taught, although the interior decoration may be somewhat subjective. More seriously, many accidents happen because users are not paying attention to changing conditions or different environments in the lab. Computers do not suffer from this lack of awareness, and by logging everything that happens in the lab and making sure that conditions are as safe as possible, experimental outcomes can only benefit. |
| FLYdom of movement | The behaviour of many organisms is strongly affected by their movement, which often gets ignored when designing experiments; animals are often fixed in place, suspended on treadmills, balls, or floating platforms, so they can be imaged while "moving", using big, bulky microscopes. This is especially true for flies, which lack the strength or size to carry even a simple imaging system. Nevertheless, these configurations are rarely satisfying as an experimental paradigm; it would be better to move the microscope. This is where we step in. The Rowlands lab is working on a high-performance, lightweight robotic fly-tracking microscope that can move to keep a fly within the field of view as it walks around. The student will be refining this instrument, improving the tracking of flies and ensuring that detailed images may still be taken regardless of how fast the fly moves. Ultimately, the instrument will be used to record neural activity in awake, behaving flies as they interact with their environment. The student who takes this project will have a strong grasp of programming, electronics and an ability to succeed despite challenging the limits of what is possible. |
| Streaming Continuous Optical Nanosecond Events (SCONE) | When high-intensity focussed ultrasound is directed at the brain, in can cause microbubbles injected into the blood stream to break through the blood-brain barrier, allowing drugs and other treatments to reach an organ that is usually carefully protected. Now obviously this has very clear medical utility, but the problem is we have no idea how the microbubble does this, and thus can't optimise the process. SCONE is a project to record an oscillating microbubble at roughly 20 million frames per second, so that when it does break through the blood-brain barrier, we can see what it is doing. SCONE requires computational reconstruction to recover the microbubble data; the student will therefore be employing advanced data recovery and modelling algorithms to address this challenge. A strong background in mathematics and / or programming would therefore be advised. |
| iPhone Imaging Improvements | Colour cameras work by filtering out around 66% of the light hitting them. Researchers in the Rowlands lab have recently developed a method for making those colour filters work with 0% loss, potentially making every camera on the planet three times as sensitive. Currently we need to explore the parameters used to design these lossless colour filters before we start commercializing / selling the design. The student on this project will be working on the design, prototyping and testing of these filters, using advanced simulation tools, nanofabrication and the analytical instruments available in the Rowlands lab. As with anything in the Rowlands lab, no specialized prior knowledge is required and all techniques can be taught. |
| Evading Export Edicts for Illegal Imaging in the Infrared | Infrared cameras can see through fog, paper and bodies, which makes them well suited to clinical microscopy. Unfortunately these cameras are "Export Restricted" - possessing them is illegal if they can image too quickly, or with too much sensitivity, or with high resolution, which makes the ones you can buy pretty much useless for bioimaging... unless you combine them. I want to build an infrared microscope that can see with unprecedented sensitivity, resolution and speed, and that hopefully doesn't get me sent to prison. The student will design and build a testbed to prototype these designs. Any skills necessary can be taught, but the student should possess a can-do attitude, and good problem-solving abilities. |
Dr Guang Yang
Profile: https://profiles.imperial.ac.uk/g.yang
Contact details: g.yang@imperial.ac.uk
| Project title | Description |
| Synthesizing Immunofluorescence Images from Bright Field Images | Immunofluorescence imaging is valuable for visualizing cellular structures but is costly and resource-intensive. Bright field imaging is more accessible, so creating immunofluorescence images from bright field images can make detailed cellular analysis more widely available. The student will need to use a generative adversarial network (GAN) trained on paired bright field and immunofluorescence images to generate synthetic fluorescence images from bright field inputs. The target is to develop a tool that accurately produces immunofluorescence images from bright field images, providing detailed cellular insights without the need for expensive staining techniques. |
| Concept and Visual guided Large Vision Language Model for Medical Report Generation | Different imaging protocols can produce varying image data even for the same type of cells. To address this, the student will develop a generative adversarial-based or diffusion-based model trained on paired images obtained from different cell painting protocols. The goal is to standardize these images, making data from diverse protocols consistent and comparable. This tool will harmonize cell painting images, enhancing the reliability and accuracy of cellular analysis without the need for uniform imaging conditions. |
| Enhancing Cell Image Segmentation through Data Harmonization | Cell image segmentation is crucial for accurate cellular analysis, but variations in imaging protocols can affect segmentation quality. The student will develop a harmonization algorithm that standardizes cell images from different protocols before segmentation. By training a machine learning model on harmonized datasets, the tool will improve segmentation accuracy across various imaging conditions, making it easier to analyze and compare cellular structures. |
| Self-supervised learning for cell imaging foundation model | Cell imaging generates vast amounts of data, but labeled datasets are often limited and costly to produce. Self-supervised learning (SSL) offers a way to leverage large amounts of unlabeled data for pretraining models, which can then be fine-tuned with limited labeled data for specific tasks. This approach can significantly improve the performance of models in various cell imaging applications, such as segmentation, classification, and feature extraction. |
Dr James Choi
Profile: https://profiles.imperial.ac.uk/j.choi
Contact details: j.choi@imperial.ac.uk
| Project title | Description |
| Visualising Sound using Machine Learning and/or Signal Processing Algorithms | Purpose. The purpose of this project is to develop a deep neural network or beam forming algorithms that can reconstruct the location of acoustic sources using multiple microphones. Motivation. In therapeutic ultrasound, a focused ultrasound transducer is used to concentrate energy to a point in the body, allowing us to noninvasively and locally manipulate tissue (tumour ablation, drug release from acoustically-active particles, etc). Our laboratory developed therapeutic ultrasound devices for delivering drugs to the brain (across the blood-brain barrier) for the treatment of brain cancers, neurodegenerative diseases, and other neurological conditions. However, the success or failure of the technique has been difficult to track as clinicians are unable to directly observe what is happening within the body. An emerging way of monitoring this procedure is with the use of microphones located around the focused ultrasound transducer. Sound generated during the procedure are captured by the microphones. We then reconstruct an image of the treated area using passive beamforming algorithms. The reconstruction of a signal source based on multiple sensor signals is broadly known as beamforming. In addition to medical imaging, it is used in underwater acoustics, astronomy, and other disciplines. The problem with many existing passive beamforming algorithms is the poor spatial resolution in the reconstruction of the sound sources. This means we can't precisely locate where the source is coming from. The purpose of this project is to develop a deep neural network and/or signal processing methods that can reconstruct an image of the treated region with better accuracy and spatial resolution. Work description. This work will involve generating training data using computer simulations on a Matlab toolbox known as k-wave. We will then train the deep neural network on PyTorch or develop fundamental signal processing algorithms. We will explore conventional neural networks such as convolutional neural networks, recurrent neural networks, and others; and, potentially, more advanced techniques, such as transformers and physics-inspired neural networks. |
| Ultra-High-Speed Video Camera at 10 Million Frames per Second | Purpose: Develop an ultra-high-speed video camera at 50 million frames per second. Motivation: Certain phenomena, such as ultrasound imaging and therapy with microbubbles (contrast agents) operate in the MHz rate. Such fast dynamics cannot be captured using traditional cameras, which operate at around 60 Hz. Commercially available cameras can reach 1 million frames per second, which is still not enough. And while a 10 million frames per second camera is available on the market for £200k, that camera cannot capture more than 256 frames (25.6µs of data). We propose a new video camera concept that could reach up to 50 million frames per second, capable of capturing nearly unlimited amounts of frames. By creating this device, we would be able to observe phenomena in biological tissue that no one has been able to observe. Outside of the domain of biomedical engineering, this camera could be used to image plasma in fusion reactors, high-speed objects in space, and other high-speed phenomena that requires incredibly high frame rates. Work: This project requires electrical engineering skills. The students would be asked to make circuits that are connected to a unique sensor array. If the analog circuit is successful, we would then require some digital electronics skillsets and optics (physics). |
| Visualising Sound - A Neural Network for Passive Acoustic Beamforming | Visualising Sound - A Neural Network for Passive Acoustic Beamforming |
| Hydrogel-based microchannels for studying acoustic cavitation | Create hydrogel-based capillary channels down to 5-10 µm in diameter. See https://onlinelibrary.wiley.com/doi/full/10.1002/smll.202301163 |
| Optimising of an ex vivo brain slice platform for studying acoustic cavitation-mediated blood-brain barrier opening | Develop methods and devices for studying mechanical phenomena in ex vivo brain slices. This may include many goals depending on the student's skillsets. For example, improve the health of the tissue and ensure that it is as close to physiological relevance as possible. Other examples include improving methods for interrogating the brain slices, such as injection of molecules directly into vessels, and increasing the throughput of these experiments. |
| Serum biomarkers of blood-brain barrier disruption induced by focused ultrasound and microbubbles. | Analyse blood samples for evidence of a change to blood-brain barrier permeability. |
| Scalable Digitisation and Readout Architecture for Large SiPM Arrays | This is electrical engineering work to create a camera using an array of SiPM sensors that can count individual photons. You'll be building the hardware the converts analog electrical signals into digital outputs. This platform can be used for high frame rate imaging (millions per second), depth imaging, and other applications. |
Dr Jun Ishihara
Profile: https://profiles.imperial.ac.uk/j.ishihara
Contact details: j.ishihara@imperial.ac.uk
| Project title | Description |
| Targeting of rare collagen for tumor immunotherapy | Cancer has high expression of rare collagen types because of an abnormality. We made antibodies for these rare collagens and try to deliver therapeutics specific to the tumours. Students who join this project needs prior cell culture experience and wet lab (molecular biology) experience. We expect to spend a long time of the day (almost every weekday) |
Dr Pedro Ballester
Profile: https://profiles.imperial.ac.uk/p.ballester
Contact details: p.ballester@imperial.ac.uk
| Project title | Description |
| Optimal design of virtual screening benchmarks from in vitro screening data | Introduction: Virtual screening (VS) has become an important source of small-molecule drug leads. A benchmark is needed to identify the VS method/s that will perform best prospectively for that therapeutic target. Benchmarks are also needed to find the optimal settings for the selected VS method/model. A VS benchmark is a library with two classes of molecules: those whose activity for the target is above a given threshold - actives as the positive class- and those with weaker or no activity at all - inactives as the negative class. Among them, a screened library is one whose molecules have been screened in the same centre and using same assay/s, e.g. the results of high-throughput screening (HTS) a compound library against the considered target. Unfortunately, HTS data have been used for VS benchmarking in a convenient yet unrealistic manner (e.g. generating benchmarks with much smaller chemical diversity than HTS). The question is how useful are these HTS-derived datasets as VS benchmarks with respect to the ground truth represented by the unadulterated HTS datasets. About the supervisor: Dr Pedro Ballester has over 17 years of experience in this research area. His last papers in this area have shown the potential of Artificial Intelligence (AI) for structure-based drug design: https://wires.onlinelibrary.wiley.com/doi/abs/10.1002/wcms.1478 https://academic.oup.com/bib/article-abstract/22/3/bbaa095/5855396 Research plan: The student will start by learning about these data types as well as existing VS benchmarks (e.g. MUV) and VS methods (e.g. USR, Smina). Then, s/he will be applying each VS method to rank the HTS-derived benchmark molecules in order to assess its performance on the associated target. This process will also be carried out for other VS methods, unadulterated HTS benchmarks and targets. The results will be employed to investigate to which extent the filters used to select molecules for a VS benchmark make it unrealistic. This is crucial for the development and selection of VS methods. About the candidate: This project is suitable for a student who is keen to learn about molecular modelling in the context of early drug design. Python programming is required. Contact: p.ballester@imperial.ac.uk |
| Optimal design of simulated-library benchmarks for virtual screening | Introduction: Virtual screening (VS) has become an important source of small-molecule drug leads. A benchmark is needed to identify the VS method/s that will perform best prospectively for that therapeutic target. Benchmarks are also needed to find the optimal settings for the selected VS method/model. A VS benchmark is a library with two classes of molecules: those whose activity for the target is above a given threshold - actives as the positive class- and those with weaker or no activity at all - inactives as the negative class. Among them, a screened library is one whose molecules have been screened in the same centre and using same assay/s, e.g. the results of high-throughput screening (HTS) a compound library against the considered target. In the absence of publicly-available HTS data, one has to resort to define a simulated-library (SimL) benchmark. Here one gathers actives from the literature and generates decoys from each of them so that these molecules collectively simulate a screened HTS library. This is the most common type of VS benchmark, where decoys are often required to have dissimilar chemical structures to that of their active to make them even less likely to be also active (activity is a rare event and hence randomly-chosen molecules are already unlikely to be active). Furthermore, an active and their property-matched decoys have similar physico-chemical properties so that they cannot be trivially discriminated by one of these properties, as this does not happen prospectively either. One popular example of such benchmarks is DUD-E (http://dude.docking.org/). The question is how useful are these SimL benchmarks for VS with respect to the ground truth represented by the entire HTS datasets (in the targets for which these are available). We expect this study to reveal which are active-to-inactive ratios and decoy schemes best anticipate performance on HTS datasets. This will be useful to decide best SimL benchmarking in absence of HTS data. About the supervisor: Dr Pedro Ballester has over 17 years of experience in this research area. His last papers in this area have shown the potential of Artificial Intelligence (AI) for structure-based drug design: https://wires.onlinelibrary.wiley.com/doi/abs/10.1002/wcms.1478 https://academic.oup.com/bib/article-abstract/22/3/bbaa095/5855396 Research plan: The student will start by learning about these data types as well as existing VS benchmarks (e.g. MUV) and VS methods (e.g. USR, Smina). Then, s/he will be applying each VS method to rank the DUD-E benchmark molecules in order to assess its performance on the associated target. This process will also be carried out for other VS methods, unadulterated HTS benchmarks and targets. The results will be employed to investigate to which extent the design factors for a retrospective VS benchmark make it unrealistic. This is crucial for the development and selection of VS methods. About the candidate: This project is suitable for a student who is keen to learn about molecular modelling in the context of early drug design. Python programming is required. Contact: p.ballester@imperial.ac.uk |
| Integrating multi-omics data from cancer cell lines with chemical structure data from molecules for enhanced pharmaco-omic modelling | Predictive models built with artificial intelligence (AI) methods are powerful tools to predict molecule-induced growth inhibition of cancer cell lines. Despite important successes, there are major challenges limiting the potential of such AI models. Some are specific to this problem (e.g. how to augment pharmaco-omic datasets in a way that improves the performance of these models). Other challenges are also found in other supervised learning problems (e.g. accurately delimiting the applicability domain of the model). This project will take a multi-task learning approach to this problem by integrating multi-omics data from cancer cell lines with chemical structure data from molecules. This will require care in acquiring and preparing omics datasets such as miRNA or mRNA expression, among others. Some introductory readings: https://doi.org/10.1371/journal.pone.0061318 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4681992/ https://doi.org/10.1093/bib/bbab312 Contact: p.ballester@imperial.ac.uk |
| A toxicity prediction toolbox based on the Therapeutic Data Commons benchmarks | Drug candidate molecules can induce many forms of toxicity in humans, often leading to the failure of a drug development programme or even the abandonment of an otherwise promising therapeutic target. Reliable computational methods capable of predicting toxicity endpoints at early stages of drug discovery are therefore of considerable importance. Recent advances in Artificial Intelligence (AI) have led to the development of highly predictive models for a wide range of toxicity-related tasks. The Therapeutic Data Commons (TDC) Toxicity Benchmarks (https://tdcommons.ai/single_pred_tasks/tox) provide a collection of standardised datasets and benchmark tasks covering multiple toxicity endpoints, alongside public leaderboards highlighting state-of-the-art AI approaches. Examples include benchmark tasks such as hERG Toxicity Prediction (https://tdcommons.ai/benchmark/admet_group/20herg), which focuses on predicting cardiotoxicity risks associated with drug molecules. This project is for students with some background in AI and Python programming. The student will: - Review the scientific literature surrounding selected toxicity prediction problems; - Investigate and explain the leading AI models used for these tasks in a clear and accessible manner; - Reproduce and evaluate published models using additional performance metrics beyond those reported in benchmark leaderboards; - Apply selected models to external molecular datasets provided by the supervisor; - Develop practical skills in running and adapting computational workflows, including Python scripts and machine learning pipelines. The project provides an opportunity to gain hands-on experience in a highly relevant area of AI-driven drug discovery while strengthening programming and computational research skills. Students will become familiar with modern toxicity prediction benchmarks, molecular datasets, model evaluation methodologies and the practical challenges associated with deploying AI models in cheminformatics research. Contact: p.ballester@imperial.ac.uk |
| Optimal selection of structure-based AI models to search for drugs leads against a target | 3D atomic-resolution model of a therapeutic target (e.g. a crystal structure of PARP1 as a target for breast cancer) can be leveraged to search for small-molecule drug leads. There is a plethora of structure-based AI models designed for this application. However, typically few of them release code that can be used on other sets of molecules. More importantly, we currently do not have any way to anticipate which of the available models to employ for a given target. The supervisor will provide targets with their structures and sets of molecules known to bind to those targets (i.e. binding molecules or binders). For each of the targets, the student will review the literature to introduce the target and find structure-based AI models for virtual screening (i.e. to discriminate between binders and non-binders in a given set of molecules). The student will evaluate the models to find out which is the most effective for the target. Some strategies to anticipate which is the best model will be evaluated (e.g. calculating the similarities between the considered targets and those included in the training set of that model). Thus, independence in running code such as Python scripts is required. This project is an opportunity to be introduced to an important drug discovery problem, the best AI models to tackle this problem and hone your programming skills. References: https://www.nature.com/articles/s41596-023-00885-w https://wires.onlinelibrary.wiley.com/doi/abs/10.1002/wcms.1478 Contact: p.ballester@imperial.ac.uk |
| Evaluating Generative AI Methods for De Novo Drug Design | De novo drug design aims to generate novel drug-like molecules directly, rather than identifying candidates from existing molecular libraries. Recent advances in Generative Artificial Intelligence (AI) have enabled the automated design of molecules that satisfy user-defined objectives such as predicted biological activity, selectivity, toxicity, novelty or synthetic accessibility. In this context, “generative†refers to AI systems that propose entirely new molecular structures optimised to meet specified design requirements. However, these requirements are typically based on computationally predicted molecular properties, which can be error-prone. For example, molecules predicted to be active against a therapeutic target may fail during in vitro validation or generated compounds may be impossible to synthesise. Consequently, understanding the practical performance, flexibility and limitations of Generative AI methods is an important and timely challenge in computational drug discovery. This project will investigate and benchmark existing Generative AI approaches for de novo drug design. The student will survey current methods and frameworks such as REINVENT (https://github.com/MolecularAI/REINVENT4). The project will evaluate these methods according to several performance indicators, including: - How rapidly a method can generate molecules predicted to reach a target activity threshold. - The scalability and computational efficiency of the generation process. - The ease with which additional objectives or constraints can be incorporated (e.g. adding a toxicity predictor or synthetic accessibility score). - The flexibility of modifying existing optimisation components, such as replacing or retraining activity prediction models. The student will gain experience in machine learning for molecular generation, reinforcement learning and optimisation strategies, cheminformatics workflows and benchmarking methodologies for AI-driven drug discovery. Contact: p.ballester@imperial.ac.uk |
| Survey and Benchmarking of Computational Toxicity Prediction Tools for Drug Discovery | Drug candidate molecules can induce many forms of toxicity in humans, often resulting in the failure of a drug development programme or even the abandonment of a therapeutic target altogether. Reliable computational methods capable of predicting toxicity risks at early stages of drug discovery are therefore of considerable importance. Over the past decade, a large number of computational approaches, including both Artificial Intelligence (AI)-based and non-AI-based cheminformatics methods, have been developed to predict specific toxicity endpoints. However, for many toxicity problems it remains unclear: - which predictive tools are currently available, - how well these methods generalise to new molecules outside their original training datasets, and - how computationally efficient they are when applied to large molecular libraries. This project is suited for students with an interest in cheminformatics, AI and computational drug discovery. Some experience with Python programming or data analysis tools would be beneficial. The student will conduct a broad literature review to identify major drug-relevant toxicity endpoints and the computational studies associated with them. For each endpoint, the student will identify publicly available prediction tools and models,determine whether these tools can be applied to user-supplied molecules, assess their accessibility, reproducibility, and usability, critically evaluate their reported predictive performance and potential limitations and compare their computational efficiency where possible. The project will make use of several important community resources and repositories, including: - Therapeutic Data Commons (TDC) for datasets, benchmarks, and leaderboards (Therapeutic Data Commons). - the associated TDC publication in Nature Chemical Biology. - ADMET-AI (ADMET-AI publication). - OCHEM toxicity modelling competitions (OCHEM Challenges). - the Polaris/Auroris benchmark repository (Auroris Benchmark Repository). - Papers With Code for identifying relevant machine learning studies and leaderboards (Papers With Code). - the Computer-Aided Drug Design (CADD) Vault (CADD Vault). - and the VLS3D directory of computational chemistry tools and databases (VLS3D Directory). This project provides an opportunity to gain experience in a highly relevant area of modern drug discovery while developing skills in literature analysis, benchmarking methodology, cheminformatics software and scientific programming. Students will become familiar with both AI and non-AI approaches for molecular toxicity prediction and gain practical experience in evaluating and running computational research tools. Contact: p.ballester@imperial.ac.uk |
| Benchmarking Tools for Molecular Property Prediction in Drug Discovery | In modern drug discovery, molecular properties such as absorption, distribution, metabolism, excretion and toxicity (ADMET), quantitative estimate of drug-likeness (QED) and synthetic accessibility (SA) are critical for prioritising promising drug leads. Numerous computational tools have been developed to predict these properties. However, it remains unclear which predictors provide the best balance between performance indicators (e.g. generalisability to unseen molecules, reproducibility and computational efficiency at large scale). This project will investigate and benchmark existing predictive tools and datasets used for molecular property prediction. The student will conduct a systematic survey of publicly available resources and frameworks, including Polaris (https://polaris-hub.github.io/polaris), TDC (https://tdcommons.ai) and MoleculeNet (https://moleculenet.org), with a focus on methods that allow investigating the best balance between performance indicators. The project will evaluate: - How well existing predictors generalise to molecules outside their original training sets. - The reproducibility of published benchmarking results. - The computational cost and scalability of these methods when applied to very large molecular libraries (potentially trillions of compounds). The student will gain experience in cheminformatics, machine learning evaluation, benchmarking methodology and high-throughput computational screening. Depending on progress and background, there may also be opportunities to implement automated benchmarking pipelines and investigate trade-offs between predictive accuracy and inference speed. Contact: p.ballester@imperial.ac.uk |
| Survey and Evaluation of Computational Methods for Molecular Target Prediction | Understanding all the macromolecular targets bound by a drug molecule is essential for elucidating its mechanism of action (MoA) in the treatment of disease. Predicting the targets of small molecules is therefore a key challenge in modern drug discovery. This problem has become increasingly important due to the growing recognition that drug efficacy, off-target effects and adverse side effects are often linked to interactions with multiple biological targets. In parallel, the resurgence of phenotypic screening approaches has led to the discovery of many active compounds whose molecular targets remain unknown. A large number of computational methods for molecular target prediction have been developed in recent years. Typically, these are ligand-based, structure-based, network-based and/or machine learning approaches. Examples of relevant studies and methodologies include: * https://pubs.rsc.org/en/content/articlehtml/2025/dd/d5dd00199d * https://pubs.acs.org/doi/10.1021/acs.jcim.6c00299 * https://pubmed.ncbi.nlm.nih.gov/31220208 Despite the large number of published methods, it remains unclear: * which tools are publicly available and operational for predicting targets of user-supplied molecules, * how these methods compare in terms of predictive performance, applicability and usability, * what their main strengths and limitations are, * and how reproducible their published retrospective validation studies are. This project is suitable for students with an interest in cheminformatics, computational biology, machine learning or drug discovery. Some background in Python programming, data analysis or computational workflows is required. The student will conduct a broad literature survey of computational target prediction methods and systematically analyse the available tools and resources. The project may involve: * identifying and categorising target prediction methodologies, * investigating publicly accessible software and web servers, * evaluating the reproducibility of published benchmarking studies, * comparing methods according to predictive performance, scalability, and ease of use, * and testing selected tools on additional molecules provided by the supervisor. This project offers an opportunity to gain experience in an important and rapidly evolving research area, while developing skills in scientific literature analysis, cheminformatics software evaluation and computational research workflows. Contact: p.ballester@imperial.ac.uk |
Professor Darryl Overby
Profile: https://profiles.imperial.ac.uk/d.overby
Contact details: d.overby@imperial.ac.uk
| Project title | Description |
| Next-Generation Explant-in-Chip: Continuous Long-Term Perfusion and Metabolic Monitoring of Tissue Explants | Maintaining viable tissue explants ex vivo requires preserving the native microenvironment, particularly the physiological delivery of oxygen and nutrients through the vascular network. We have developed a novel explant-in-chip platform that achieves intravascular perfusion, forcing flow through native blood vessels to deliver oxygen and nutrients directly to cells via their natural vascular route, thereby maintaining tissue viability and function. However, current implementations are limited by finite reservoir capacity, restricting experiments to several hours. Extending perfusion duration to multiple days would enable longitudinal studies of tissue metabolism, drug pharmacokinetics, and therapeutic responses while continuously monitoring metabolic status to verify tissue health. This project offers multiple engineering challenges. One challenge involves developing a continuous recirculation system to enable multi-day perfusion. The current perfusion design uses supply and waste reservoirs that deplete within hours. Building on preliminary flow rectifier modules developed in the laboratory, students will optimise a "microfluidic diode" circuit analogous to an AC/DC electrical bridge rectifier that converts bidirectional flow into unidirectional perfusion across the explant. Importantly, this design will enable indefinite unidirectional perfusion through explant vasculature despite bidirectional reservoir exchange. A second challenge involves integrating electrochemical biosensors for real-time metabolic monitoring. Collaborating with Prof. Martyn Boutelle's laboratory, who have pioneered rapid-response glucose and lactate biosensors, students will incorporate these sensors into the perfusion loop. This integration will provide continuous readouts of tissue metabolic activity, enabling immediate detection of viability changes, optimisation of perfusion conditions, and real-time assessment of drug-induced metabolic perturbations. Preliminary data demonstrate stable sensor performance in perfusion media, with rapid response times suitable for closed-loop control. These engineering advances will enable unprecedented long-term maintenance of tissue explants with real-time metabolic feedback, advancing applications in drug testing, disease modelling, and personalised medicine. Success will be measured by demonstrating multi-day tissue viability through continuous perfusion and/or real-time glucose and lactate monitoring in liver explants. |
| Connecting living and engineered systems via the vasculature using microfluidics | There is growing demand for new approach methodologies (NAMs) using human-relevant models to reduce reliance on animal testing and better predict how patients respond to drugs. Tissue slices keep the native architecture and cell diversity of the native tissue, but without blood flow they starve beyond their outer layers. Our laboratory has developed a novel design that drives flow through the tissue's own vasculature. This allows living tissue to be connected by flow to engineered systems that support its viability and function and allow drug responses to be tested. The project forms part of INTREPID, a multi-partner programme developing live human tumour explants as alternatives to animal models. The project aims to (i) engineer the tissue–device interface for optimal perfusion through the tissue vasculature, (ii) demonstrate robust preservation of tissue viability over time, and (iii) demonstrate a measurable drug response. Work will focus on murine tissue, with access to patient-derived explants through INTREPID. The project can take either of two directions, suited to one or more students. Experimental: The student will fabricate candidate designs using a high-resolution 3D printer and test them on tissue slices. Readouts include perfusion flow rate, hydraulic resistance through the tissue, tracer delivery and tissue viability. Through our collaboration with the Institute of Cancer Research, microbubble ultrasound imaging will show which vessels are perfused. Computational: The student will build finite element models of the tissue–device interface under perfusion. These will identify the design features that best seal the tissue while keeping its vessels open, guiding the next generation of devices. The student will gain hands-on training in device design, precision 3D printing, perfusion and tissue biology, or in computational mechanics. They will work within a team spanning engineering, cancer biology and imaging, on a platform heading towards clinical and industrial use. |
| A bio-inspired approach to interface living tissue slices with microfluidics and microscopy | There is growing demand for new approach methodologies (NAMs) that use human-relevant models that reduce reliance on animal testing and better predict how patients respond to drugs. Perfusing a living tissue specimen, rather than bathing it in medium, is what keeps it alive and allows drugs and tracers to be delivered the way they arrive in the body. The challenge is sealing. Fluid will follow take the easiest path, so unless the edges of the specimen are well sealed, medium simply escapes around the tissue instead of passing through it. Existing seals typically clamp or glue the specimen in place, which damages delicate tissue, are irreversible which limits further analysis, and obstructs the optical access needed to watch what happens inside. This project will develop an alternative inspired by the iris of the eye. Namely, a flexible aperture that closes gently around a thin tissue specimen to grip and seal its edge and can be reopened to release the tissue unharmed. The project aims to (i) develop an aperture-based seal that accommodates specimens of varying size, shape and stiffness, (ii) demonstrate that it directs flow through the tissue rather than around it, and (iii) demonstrate that the sealed specimen can be imaged live under flow. The design will be tested on thin-cut murine tissue, such as liver, tumours or the outflow tissues of the eye as cases of particular interest. The student will design and fabricate prototype apertures and housing devices in flexible polymers using soft lithography and high-resolution 3D printing, exploring how aperture geometry, membrane thickness and actuation determine the grip on the tissue and the associated mechanical stresses. Performance will be assessed by measuring flow rate and hydraulic resistance through the specimen, by quantifying leak around the seal using fluorescent tracers, and by confirming that the tissue remains viable and undamaged after release. By applying microfluidic approaches, the housing can be made thin enough to be compatible with light microscopy. The student will gain training in device design, microfabrication, soft material mechanics, perfusion measurement and live tissue imaging. They will join a team developing perfusion technologies for tissue explants, on a device intended for use across many tissue types. |
| Microvascular Plumbing: Linking vascular architecture to flow resistance in engineered microvascular networks | There is growing demand for new approach methodologies (NAMs) that use human-relevant models to reduce reliance on animal testing and better predict how patients respond to drugs. A major advance has been the ability to grow living, perfusable blood vessel networks inside microfluidic chips, where endothelial cells suspended in a hydrogel self-assemble into a branching network that can be connected to flow. These vascularised models are increasingly used to study drug delivery, tumour invasion and vascular disease. What is almost never measured, however, is how well they actually work as plumbing. Hydraulic resistance is the property that determines how much flow a network carries and how much pressure the tissue sees, and it is set by the architecture of the network. Namely, how many vessels there are, how wide, how well connected, and how leaky. Our laboratory grows vascular networks within microfluidic devices and has built a measurement platform, iPerfusion, capable of resolving these resistances precisely. This project will examine the relationship between the structure of engineered microvascular networks to their function. The project aims to (i) establish reliable measurement of hydraulic resistance in engineered microvascular networks, (ii) determine how resistance relates to the architecture of the network as seen by microscopy, and (iii) determine how supporting mural cells influence the structure-function relationship. The student will culture human umbilical vein endothelial cells in fibrin hydrogels within three-channel microfluidic chips, a widely used format in which networks form over several days and connect to flanking channels for perfusion. Supporting cells, such as mesenchymal stromal cells, fibroblasts or smooth muscle cells, will be co-cultured in the gel to promote vessel maturation, and the student will compare networks grown with and without support cells. Resistance will be measured using iPerfusion, and the same chips imaged by confocal microscopy to extract vessel density, diameter, branching and connectivity. Relating the two will reveal which architectural features most strongly influence flow, and whether resistance can serve as a rapid functional readout of network quality in place of imaging. If time allows, the platform opens onto questions the laboratory is keen to pursue. For example, whether resistance changes with perfusion pressure (e.g., flow homeostasis via shear-induced nitric oxide signalling) and how the resistance responds to vasoactive or anti-angiogenic drugs. The student will gain training in microfluidic cell culture, organ-on-chip approaches, vascular biology, precision flow and pressure measurement, confocal imaging and quantitative image analysis. They will join a team developing perfusion technologies for living tissue, on a model system with broad relevance across vascular research and drug testing. |
Professor Mengxing Tang
Profile: https://profiles.imperial.ac.uk/mengxing.tang
Contact details: mengxing.tang@imperial.ac.uk
| Project title | Description |
| Depth-Resolved Blood Flow Velocity Estimation for Transcranial Doppler Ultrasound | Background Accurate estimation of cerebral blood flow velocity is critical for diagnosing cerebrovascular conditions. Transcranial doppler (TCD) uses ultrasound waves to measure cerebral blood flow velocity in the major intracranial arteries, which can be an indicator of increased stroke risk for children with sickle cell disease [1]. TCD provides blood flow spectra based on a single element transducer or color-coded blood flow based on a phase/linear probe. Both requires three-dimensional knowledge of cerebrovascular anatomy and its variations to localise targeted arteries as well as precise probe positioning for tracking/monitoring. Besides, the interpretation of the blood flow spectra for diagnosis and treatment demands professional training and practical experience, making the TCD study highly operator dependent. Recently, volumetric ultrasound imaging has been developed and explored for transcranial blood flow imaging [2], enabling comprehensive visualisation of intracranial arteries with minimised manual adjustments. In this project, a deep learning framework is proposed to directly estimate blood flow velocity from volumetric doppler ultrasound data (in human). Aims and methods: A deep learning model will be trained to estimate blood flow velocity using doppler signals. Different data preprocessing including IQ, phase, and radiofrequency will be explored. The proposed method will be compared against conventional doppler processing (FFT-based and autocorrelation methods). The proposed framework will be validated using simulation, flow experiments, and in vivo human volunteer data. Skills: Knowledge of ultrasound simulation software such as Field II and k-Wave; practical experience of deep learning (Keras, PyTorch) is preferable but not mandatory References: [1] R. Adams et al., “The Use of Transcranial Ultrasonography to Predict Stroke in Sickle Cell Disease,” New England Journal of Medicine, vol. 326, no. 9, pp. 605–610, Feb. 1992, doi: 10.1056/NEJM199202273260905. [2] P. Xing et al., “3D ultrasound localization microscopy of the nonhuman primate brain,” eBioMedicine, vol. 111, Jan. 2025, doi: 10.1016/j.ebiom.2024.105457. |
Professor Rylie Green
Profile: https://profiles.imperial.ac.uk/rylie.green
Contact details: rylie.green@imperial.ac.uk
| Project title | Description |
| Implanted device development for targeted in-tumour delivery of chemotherapeutics | Design and prototyping of device for delivery of electronic chemotherapy. Understanding fabrication and drug loading interactions and subsequent impact on delivery into brain tumour tissues. |
| Extended-Release Intraperitoneal Chemotherapy via PLGA-Loaded Injectable Hydrogel for Enhanced Treatment of Peritoneal Metastasis | Extended-Release Intraperitoneal Chemotherapy via PLGA-Loaded Injectable Hydrogel for Enhanced Treatment of Peritoneal Metastasis |