{"id":"https://openalex.org/G589212349","display_name":"Cambridge, Mathematical Genomics and Medicine.","description":"The Tufts Center of the Study for Drug Development estimates the cost of developing a drug and bringing it to market is $2.6 billion, and advances in the fields of machine learning and computational statistics have the promise to decrease these costs significantly and have a significant impact in improving global human healthcare. Genome Wide Association Studies (GWAS) have provided a comprehensive map of genotype to high level phenotype associations, for example, diseases like schizophrenia or inflammatory bowel disease. A key challenge is to understand the lower level molecular processes and pathways that cause diseases, which will enable better drug development. These phenotypes are high-dimensional and are increasingly being captured using single-cell assays and high-content imaging. It is important to understand cell trait variation across natural genetic differences and engineered variants, thereby elucidating the cellular consequences of disease mutations. Mapping low-level cellular phenotypes often have much higher heritability when compared to complex high-level phenotypes. Focussing on molecular phenotypes evades the missing heritability problem that plagues GWAS. A major step to release the potential of cellular screens is the automated analysis through processing of image data. These datasets contain many images, with high- dimensional phenotypes, and a high noise to signal ratio. Modern machine learning methods are well suited to this problem, using deep learning, making use of computational advances of consumer grade high-performance graphical processing units (GPUs). This project will design the best methods to exploit cellular images in a range of contexts: \\- Develop deep learning frameworks for feature extraction from high throughput microscopy data. \\- Use the features to investigate the link between genetic variation and cell trait variability in existing human induced pluripotent stem cell images, and new CRISPR/Cas9 knockout lines. \\- Extend the models to integrate other omics modalities and existing data. Machine Learning for high-content imaging In understanding how individual cells work, their characteristics must be measured. A standard way to do this is using high-throughput microscopy, which is a rich source of quantitative data, with thousands of pixel values measured for every cell. The main problem in making use of this large-scale information is extracting meaningful features. There are many existing software packages that are able to generate numbers from each cell image, ranging from area and shape characteristics to values for Zernike features and Gabor filters, but these hundreds of numbers rarely correspond directly to a biologically interpretable signal. Alternatively, a desired biological characteristic can be modelled and quantified from the images, but this requires extensive feature engineering. A promising alternative to the abundant local and scarce broad global image features is to extract them from deep neural networks. These models are able to make use of large-scale data to learn spatial correlations that are not easily inferred from standard parametric models. In the last years, this class of methods has outperformed alternatives on most image processing tasks, ranging from object detection to semantic embedding. I will explore a collaboration available to me with AstraZeneca, on using image analysis to optimise their virtual screening process. Once a potential drug target or mode of action for a disease is identified, assays are designed, and massively parallel high-throughput drug screens are conducted to identify compounds that successfully interfere with this drug target and achieve a positive assay result. I aim to detect automatically and at an early stage the outcome of a compound for a given trial from an intermediate image that is taken of the assay, reducing the number of false positive results that are carried forward for expensive and time-consuming counter screens, and potentially even obviating the need for them. Related work has shown it is possible to characterise complex cellular phenotypes [1] using deep learning [2], and has also discussed its use in virtual screens [3]. Linking genetic variation to cellular phenotypes The methods established in Aim 1 allow extracting interesting features from many unrelated individuals in the HIPSCI [4] project. The next natural problem is linking genetic variation in them to these features after genotyping and identifying causal mutations, as well as sharing of the underlying mechanisms via colocalization approaches. I am interested in genetically linking cellular traits, and methods that have been developed for finding causal SNP configurations after genotyping. I intend to use stochastic search algorithms as proposed by Brenner et al [5] to find causal configurations on large candidate gene-set ranges. I\u2019m interested in how these methods can be optimised for this use case where I anticipate heritability to be higher. Furthermore, I\u2019m interested in using colocalization methods [6] to investigate the existence of shared genetic basis between phenotypes. In my rotation report, I implemented modern day techniques which cross-multiply independently generated posteriors, and I will look to optimise these techniques for cellular traits. I am also interested in extending these further to detect whether traits in question share a genetic basis only partially. Here I will be interacting with Oliver Stegle and his group which has extensive expertise in statistical genetics. In addition, conducting CRISPR screens with Leopold Parts\u2019 group will allow me to investigate the cellular effects of gene knockouts. Finally, collaborating with drug companies like AstraZeneca will give me the opportunity to investigate the cellular consequences of administering drug compounds. Integrative methods to identify drug targets Given the ability to extract useful cellular features from images, and to model their sources of variability, it is possible to attack the central problem in cell biology - forming a mechanistic chain from genotype to high-level phenotypes via cellular traits. I aim to integrate disease GWAS summary statistics, along with intermediary causal layers like eQTL studies, to look for shared cellular trait QTLs and understand disease pathways at a more granular level. I want to be able to decipher whether genetic mutations impact diseases or traits directly, or whether they do so indirectly by impacting gene expression and other cellular traits. Recently published methods [9] have integrated GWAS summary data with eQTL studies and use mendelian randomisation to account for confounding non-genetic variables. In this way, variants can be better classified as causal, pleiotropic or linked. These questions are important to address in minimising false positive leads and optimising the interface between highlighting potential causal SNPs, and further investigation with functional knockouts, and I intend to extend these methods by incorporating cellular traits. In this way, I aim to build causal models to suggest drug targets. Since this work integrates many different data-layers, I\u2019m interested in investigating whether they can be used as inputs in more complex models. Working as part of Open Targets, a collaboration between the Wellcome Trust Sanger Institute, the European Bioinformatics Institute, GlaxoSmithKline and Biogen, I will be able to collaborate between research and industry to work on the most pressing problems in drug development. Conclusion In conclusion, I\u2019m interested making use of cellular microscopy images, a data types that is increasingly prevalent in understanding cellular systems. I intend to extract features from these images, model them, and link these models to genomic data to understand complex traits and aid drug discovery. I will use available data from HIPSCI, AstraZeneca, novel CRISPR knockout data as part of Leopold Parts\u2019 group, and publically available omics data from Open Targets. This field is set to explode with the advent of high-throughput cellular screens and gene-editing and I look forward to developing the next-generation methods to intelligently handle the volume of data output.","funder_award_id":"109083/Z/15/Z","funder":{"id":"https://openalex.org/F4320311904","display_name":"Wellcome Trust","doi":"10.13039/100010269"},"funded_outputs":["https://openalex.org/W2769569970","https://openalex.org/W2888300707"],"funded_outputs_count":2,"parent_awards":[],"sub_awards":[],"sub_awards_count":0,"amount":164360.0,"currency":"GBP","amount_usd":248826.81,"amount_usd_basis":"daily_rate","recipients":[{"id":"https://openalex.org/I241749","display_name":"University of Cambridge","ror":"https://ror.org/013meh722","country_code":"GB","roles":["lead"],"share":null,"basis":"affiliation_matcher","asserted_by":"funder"}],"recipient_countries":[{"country_code":"GB","share":1.0,"amount_usd":248826.81,"basis":"organisation_location"}],"recipient_regions":[],"research_scope":{"is_research":true,"basis":"funder_default","tier":1},"funding_type":"fellowship","funder_scheme":"PhD Studentship (Basic)","start_date":"2015-10-01","end_date":"2020-06-30","start_year":2015,"end_year":2020,"landing_page_url":"https://wellcome.org/grant-funding/people-and-projects/grants-awarded?q=109083/Z/15/Z","doi":null,"provenance":"wellcome_trust","lead_investigator":{"given_name":"Mr William","family_name":"Jones","orcid":null,"role_start":null,"affiliation":{"name":"University of Cambridge","country":"United Kingdom","ids":null}},"co_lead_investigator":null,"investigators":null,"works_api_url":"https://api.openalex.org/works?filter=awards.id:G589212349","primary_topic":null,"topics":null,"institution_awarded":[{"id":"https://openalex.org/I241749","display_name":"University of Cambridge","ror":"https://ror.org/013meh722","country_code":"GB","type":"education","lineage":["https://openalex.org/I241749"]}],"updated_date":"2026-11-03T03:01:00.000Z","created_date":"2026-07-31T15:41:50.000Z"}