Projects available

E1: Target identification from multi-omics data using systems biology and machine-learning approaches

The aim of this project is to develop robust explainable AI methodologies to mine large multi-omics perturbation datasets for novel mechanistic insights. In particular, we are interested in developing supervised learning models that predict the effect of chemical perturbation on individual genes and proteins. Using machine-learning approaches, you will capture omics information from chemical perturbations resources (e.g., Connectivity Map) and disease-induced perturbations to understand fundamental biological mechanisms.  

We are seeking a highly motivated students who are passionate about contributing to biological knowledge through the application of AI to large biomolecular data sets. The ideal candidates will have a grounding/interest in both molecular biology and data science – this could be through a Masters degree in a subject such as bioinformatics, or alternatively you may have a first class degree in computer science followed by bioscience experience, or vice versa. You will be confident in coding in Python, with some experience of data wrangling, statistics and machine learning. 

Exscientia is an AI-driven pharmatech company committed to discovering, designing and developing the best possible drugs in the fastest and most effective manner. You will work closely with world-class scientists at Exscientia throughout the project. You will have access to Target Analysis, Discovery and AI technology teams for advice and support during your stay at Exscientia. Within QMUL, you will work at Professor Conrad Bessant’s lab and benefit from his team’s expertise in both AI and multi-omics data analysis. 

Supervisors:
Mani Mudaliar, Target Analyst Director, Exscientia 
Conrad Bessant Professor of Bioinformatics, School of Biological and Behavourial Sciences, QMUL 

Project Partner:
Exscientia

E2: Target/Biomarker selection using systems networks and decision theory

Methods for automated target and biomarker identification typically involve ranking genes and proteins according to indication relevance criteria. However, many age-related western world diseases bear the hallmarks of complex heterogeneous conditions that vary along a spectrum of disease severity. They frequently involve multiple systems and pathways. Hence selection pools comprise many hundreds to thousands of genes. Selecting the best targets and stratification biomarkers from a large and diverse pool presents a significant challenge to drug discovery. Historically, selections have been manual or at best semi-automated. This means it is difficult to avoid bias in such decision making tasks. Decision theory is a branch of Artificial Intelligence that can be applied to help resolve such complexities. Automated reasoning and argumentation theory are two appropriate branches of machine learning well suited to the problem. The aim of this project is to fully automate target and biomarker selection processes using AI methods to make unbiased decisions from a network of information relationships.  

We are seeking a highly motivated students who are passionate about contributing to biological knowledge through the application of AI to large biomolecular data sets. The ideal candidates will have a grounding/interest in both molecular biology and data science – this could be through a Masters degree in a subject such as bioinformatics, or alternatively you may have a first class degree in computer science followed by bioscience experience, or vice versa. You will be confident in coding in Python, with some experience of data wrangling, statistics and machine learning. 

Exscientia is an AI-driven pharmatech company committed to discovering, designing and developing the best possible drugs in the fastest and most effective manner. You will work closely with world-class scientists at Exscientia, and will be based in the Biomarker and Targets Data Science Team, benefiting from their experience in AI and network systems biology methods applied to drug discovery. 

Supervisors:
Anna Lobley - Assoc. Director Biological Data Science, Exscientia

Claudia Cabrera - Lecturer in Bioinformatics, William Harvey Research Institute, QMUL

Arkaitz Zubiaga - Senior Lecturer, School of Electronic Engineering and Computer Science, QMUL

Project Partner:
Exscientia 

F1: Predicting metastasis and drug response in skin cancer using multimodal deep learning

Cutaneous squamous cell carcinoma (cSCC) is the most common skin cancer with metastatic potential globally. Surgery and radiotherapy effectively treat most cSCCs, but disease burden is significant, with an estimated 45-50,000 new cases and an incidence increasing by 5% per year in the UK. Approximately 2-3% of primary cSCC will metastasise with poor prognosis, leading to one-quarter of all skin cancer-related deaths. Being able to identify which cSCC are at high risk of recurrence or metastasis allows us to make appropriate management decisions regarding adjuvant treatment and follow-up; for example, Cemiplimab, an FDA-approved immune checkpoint inhibitor, is more likely to induced a response in advanced high-risk cSCC. However, current risk stratification tools all have significant limitations and poor performances. Thus, the accurate and reliable identification of primary cSCC at risk of locoregional recurrence and metastasis is currently a major unmet need in skin cancer care. 

We aim to create a deep learning tool, integrating digital pathology and multi-omics data, to predict the prognosis of cSCC. 

We have recently generated whole transcriptomics, whole-exome and targeted sequencing data from 150 non-metastatic and 85 metastasising cSCC patients. We have also generated the matched scanned H&E-stained whole slide images. 

We will first analyse transcriptomics and mutation data together to develop a multi-omic signature to identify high-risk cSCC comparing the two clinical groups. A range of state-of-the-art machine learning (ML) classification tools will be tested, and the best model with the best performing set of molecular events will be determined. Next, we will integrate histology and molecular data to develop a joint image-omic prognostic model using multimodal deep learning. 

This project will be of great interest to students who have quantitative background, such as Bioinformatics, Genetics/Genomics and STEM subjects. Students will be trained with cutting-edge deep-learning techniques to analyse real world patient data. 

Supervisors:
Jun Wang - Senior Lecturer in Bioinformatics, Barts Cancer Institute, QMUL 
Irene Leigh - Professor of Cellular and Molecular Medicine, Dean for Global Engagement for Medicine and Dentistry, QMUL
Catherine Harwood - Clinical Professor of Dermato-Oncology, QMUL

Project Partner:  
Sanofi 

H1: Cheminformatics and Machine Learning approaches for GPCR Computer-Aided Drug Design  

This project will focus on the development and application of cheminformatics, machine learning (ML), and Artificial Intelligence (AI) approaches for Computer-Aided Drug Design (CADD) for G Protein Coupled Receptors (GPCRs), the largest family of cell signalling transmembrane proteins that can be modulated by a plethora of chemical compounds. This project will use the information available in many heterogeneous types of protein-ligand interaction data to develop models that enable the design of efficacious therapeutic compounds targeting GPCRs. The project will make extensive use of public bioactivity data drawn both from the literature and patents, as well as experimentally determined structures of GPCR-ligand complexes. 

You will use a variety of computational chemistry and cheminformatics techniques such as similarity assessment using 2D and 3D approaches, de novo design, quantitative structure-activity relationships for property prediction, molecular interaction fields, and protein-ligand docking. You will work on the development, evaluation, and optimization of novel approaches augmenting with experimental GPCR structural biology, chemical, and pharmacological data and state-of-the-art AI and ML techniques, including deep generative models, convolutional neural network models, and reinforcement learning. The ultimate goal will be techniques and approaches that can be applied to GPCR drug discovery projects as part of the Design-Make-Test-Analyse cycle. 

We are seeking candidates with keen interest in applying computational tools to address complex chemical and biological questions. Applicants should hold a Master's degree in Chemistry or a related discipline, and should be familiar with at least one scientific programming language. Experience in computational chemistry, cheminformatics, and/or computer-aided drug discovery is desirable. The position requires excellent inter-personal, oral and written communication skills in a collaborative, interdisciplinary biotech drug discovery environment. 

Note: A computer science diploma without knowledge of chemistry will be insufficient to efficiently perform this PhD project. 

Supervisors: 
Noel O’Boyle – Principal Scientist at Sosei Heptares 
Chris de Graaf – Director, Head of Computational Chemistry at Sosei Heptares 
Arianna Fornili - Senior Lecturer in Computational Organic Chemistry, School of Physical and Chemical Sciences, QMUL 

Project Partner:
Sosei Heptares

M1: Development of AI framework for iPSC image analysis and integration with transcriptomics  

The Draviam lab aims to understand the molecular principles that govern cell division and the consequence of its failure when cells transition between states during differentiation. Together with MSD, the project aims to focus on developing AI methods to track iPSC (induced Pluripotent Stem Cells) differentiation to neural precursors, astrocytes and neurons. During the process of differentiation, quiescence and senescence are two cell dormancy states with distinct cell fates and transcriptomic statuses. However, these two states of dormancy have similar nuclear shape and size presentation (in images) which we aim to separate by developing a DL-based image analysis framework that tracks their behaviour through time. The ideal candidate is expected to have a strong background in image analysis, development of ML/DL methods, a keen interest in biosciences and provable experience in teamwork. 

Supervisors:  
Viji Draviam - Professor of Quantitative Cell and Molecular Biology and Director of Industrial Innovation, School of Biological and Behavioural Sciences, QMUL  
Victor Neduva - Senior Director, Genome and Biomarker Sciences, MSD  

Project Partner:  
MSD 

S1: Machine learning for structure-based drug & biologic discovery 

Our work is in in silico structure-based drug and biologic discovery. We shall provide platforms that will optimise exploitation of the dynamic properties of both protein targets and of the sequences of designed biologic drugs. This work builds on research in the Fornili group which shows supervised machine learning outperforms traditional pattern matching approaches used for categorising binding pockets suitable for binding small molecule drugs. The function of all proteins including G-protein coupled receptors, enzymes, ion-channels, and cytoskeletal proteins depend on flexibility. This flexibility can create new pockets and crevices suitable for binding small molecule drugs and biologics. Supported by evidence from crystallographic, NMR and cryo-EM studies, conformational differences accessed by dynamics can vary from small scale rotamer changes to the ordering of intrinsically disordered regions on the larger scale.

The first part of this work involves the application of supervised machine learning methods to the selection of representative protein conformations from ensembles generated from molecular simulation. The deliverable will be an open-access "pocket classifier". In the second part of this work, you will be exploring ensemble dynamics of the variable regions of single chain antibodies and of RNA aptamers so that structure and dynamics may be efficiently included in in silico sequence selection of potent inhibitors. Drug companies have for many years explored small molecules, but the full potential of single chain antibodies and of aptamers/synthetic aptamers has yet to be realised. The deliverable will be new machine learning-based software ideally underpinnig consultancy or a spin-off in which the student can choose to be a key player. 

Supervisors: 
Richard Pickersgill - Head of the School for Biological and Behavioural Sciences, and Professor of Structural Biology, QMUL 
Arianna Fornili - Senior Lecturer in Computational Organic Chemistry, School of Physical and Chemical Sciences, QMUL 

Project Partner:  
Evotec (UK) Ltd 

S2: Modelling the phospho-regulation of cell cycle control

This project aims to create a mathematical model that predicts how changes in phosphorylation of proteins drives cell division. Cell division is a prerequisite for all life on earth; from bacteria to humans, cells division allows organisms to reproduce and creates specialised tissues and organs in multicellular organisms. Cell division during adult life regenerates damaged tissues but, when unregulated, is also the cause of cancer. The master regulator of cell division is a single family of kinases, called cyclin-dependent kinases (CDKs). CDK activity is both necessary and sufficient to drive cell division in model systems and is often elevated in cancers. However, understanding the mechanism of CDK action is complex, since CDK phosphorylates many hundreds of substrate proteins in the cell, which play diverse roles. Additionally, counter-acting phosphatases remove the CDK-induced phosphates from substrate proteins in a highly dynamic process. How do we understand the role of CDK at specific substrates and then build up a picture of how CDK controls cell division globally? 

We aim to use novel biological data to model how the activity of CDK drives cell cycle progression, initially on a small scale, but building a framework that can be applied globally to the thousands of proteins in the cell. We have used artificial interactions of CDK (and the counteracting phosphatases) to every protein in budding yeast (a canonical eukaryotic model of cell division), to identify functional effects of CDK upon the cell cycle. These data give a global picture of the effects of CDK on different cellular systems that are important for driving cell cycle progression. This project will build a logical model that combines existing knowledge about the action of CDK with our synthetic interaction data to predict the function of CDK. We will initial restrict our model to approximately 30-60 key proteins that are involved in the final stages of cell division (mitosis), when CDK activity peaks.  

The model will make predictions that can be tested experimentally – either by creating phospho-site mutants or manipulating the activity of kinases or phosphatases. We will use high-throughput robotic handling of yeast colonies and assessment of their cell division rate – in collaboration with Singer Instruments Ltd. The model will identify key regulatory phosphorylation events and thus potential drug targets for regulating cell division. 

Supervisors:  
Oliver Severn - Research Manager, Singer Instruments Ltd 
Peter Thorpe - Senior Lecturer in Biochemistry, School for Biological and Behavioural Sciences, QMUL 
Conrad Bessant - Professor of Bioinformatics, School for Biological and Behavioural Sciences, QMUL 

Project Partner:  
Singer Instruments Ltd