Innovation Placements 2023

Large Deep Generative Graph Models for Biomedicine

Duration: Six months.
Host organisation: Epistemic AI
Location: Remote
Point of contact:  Stefano Pacifico, Co-founder & CEO
linkedin.com/in/stefanopacifico

Biomedical knowledge is vast, interconnected and growing each day and advances in the Science are driven by drawing upon this plethora of information.  For example, we might be able to infer that a drug that was originally designed to tackle depression is also effective at treating  neuralgia.  

Here at Epistemic AI  we encode this information as a Knowledge Graph (KG), containing billions of entries.  This project addresses the task of taking our KG (or a part of it) and using deep learning to represent it as graph embeddings.   Once we have the right representation, we can go on to make discoveries, uncovering connections that were previously hidden to the community or are even novel.  We are agnostic regarding programming languages and frameworks (Python and PyTorch would be good candidates).  No biomedical knowledge is required.  Instead, some machine learning and data handling skills would be useful, within a Linux environment.  The project will be heavily hands-on and practical.

The idea of the project is to create embeddings over time series data, where the data is appropriately released clinical records for HIV patients.  With these embeddings in hand, we will then be in a position to start reasoning about patients, both within and across the data sets. For a given person, with a record of treatments and symptoms, their time series (temporal embeddings) could be inspected and clinical events might be discovered. By repeating this for cohort of patients, it becomes possible to look into more generalised patterns, allowing  clinicians to draw medical inferences about patients. Here the IP would be the model itself, but the problem setting and exploration would potentially lead to a paper. 

There is considerable flexibility with this project and we expect that it will evolve over time.  Outcomes of the project will be research and development within a startup environment, software engineering skills, presentation of ideas and discussions and collaborations with subject matter experts here at EAI.  We might even create new products and make Scientific discoveries!

Further, we would welcome any candidate with independent project proposals for Epistemic AI, in addition to the current description.

Core competences

• Python
• PyTorch (desirable)
• Machine learning and data handling skills

Personal attributes and qualities

• Excellent communication skills 
• Motivated team player
• Practical and hands-on approach to problem solving and projects

Beneficial but not essential

• Understanding of the pharmaceutical industry

About Epistemic AI (EAI)

Epistemic AI was founded in 2018 to provide efficient and easily accessible AI solutions for the life sciences. The state-of-the-art AI platform eliminates common technology barriers in research and enables rapid knowledge discovery. The Epistemic AI platform effectively unlocks silos of information that can potentially deliver better cures. Epistemic AI works with academia, foundations, and biopharma companies to help advance their efforts in R&D, drug discovery, clinical trials, and commercialization. To learn more, visit epistemic.ai, or contact at info@epistemic.ai.

Deadline: 5pm on Friday 14th July

Innovation Placement 1 with the NHS

In collaboration with the NHS, DiRAC is pleased to invite applications for a 6-month Innovation Placement focussing on Expansion of the use of Virtual Wards

Host organisation: Royal Free London (RFL) NHS Foundation Trust
Location: IIA operates a hybrid working policy, with up to two days a week office based, however, fully remote would be considered with regular, pre-defined F2F meetings. As per RFL policy, employees cannot work outside the UK.  

Background:

A perfect storm of pressure has meant that urgent and emergency services have been through the most testing time in NHS history. Problems discharging patients to the most appropriate care settings, alongside the demands of flu and COVID peaking together, have contributed to hospital occupancy reaching record levels. Nationally, 19 out of 20 beds are occupied; up to 14,000 beds are occupied by someone who is clinically ready to leave. This congestion, among other things, has meant patient ‘flow’ through hospitals has been slower.  

A key component of NHS England’s Urgent and Emergency Care (UEC) recovery plan is the use of virtual wards. Virtual wards bridge the gap between hospitals and patients’ homes. They combine technology and face-to-face provision to allow hospital-level care including diagnostics and treatment, whilst at the same time allowing people to remain at home with family or carers where they are more comfortable. Virtual wards can both help get people home earlier and help avoid admissions in the first place. 

The evidence base for virtual wards is growing, with clinical evidence to show that in select situations, virtual wards are a safe and efficient alternative to NHS bedded care, particularly for patients living with frailty. 

Whilst NHS England is currently focusing on increasing capacity of virtual wards, with a long term ambition on reaching 40-50 virtual wards per 100,000 people, there is also an acknowledgement that in addition to increasing capacity, there needs to be a focus on increasing the utilisation of the virtual ward capacity that we already have. Their current short-term ambition is to increase the utilisation of virtual wards from around 65% to 80% by September 2023. 

The Project

The aim of this project is to help guide national practice by providing insight into how RFL can safely improve the utilisation of virtual wards by evaluating to understand: 

• How are we currently using virtual wards? What types of patients are on them and are the attributes of that admission that lead to the referral to virtual wards? 
• What types of patients have good outcomes on virtual wards? What types of patients are not suitable for virtual wards e.g. unsuitable home arrangements, falls risk? 
• What types of clinical activity (management of falls, delirium, end of life) needs to be provided on frailty wards. This understanding will help support the virtual ward referral process. 
• What are the benefits to patients on a virtual pathway? Benefits may include reduction in readmissions or reduction in harm associated with hospital admissions such as deconditioning, pressure sores, and hospital acquired infections. 
• What are the system-wide impacts of increased utilisation of virtual wards? 

At RFL there is the opportunity to work on both established frailty virtual wards, and on the newly created heart failure virtual wards. The virtual heart failure ward is an extension of the 1^st Cerner (Electronic Health Record system used at RFL) Digital Heart Failure pathway in the UK² and currently holds over 31k patients’ data in digital format. 

The project is designed to be accessible to candidates without healthcare data experience. Candidates will undertake training and shadowing sessions to understand how clinical data is generated and how errors can occur. The project will assess current practices, clean and organize data, and build a pipeline using data stewardship and data science. Statistical methods will be used to identify the target population and compare classical statistics approaches to machine learning. The project will also compare the impact of different data inputs and aims to produce work suitable for publication in peer-reviewed journals.

The candidate will be expected to present their findings to both data and clinical groups across RFL, including: 

• Virtual Hospital Oversight Group 
• Data steering group 

Candidate profile

Candidates will be expected to apply the tools and techniques they have learnt or used during their studies from areas with an advanced approach to data science, into an applied healthcare project.

A suitable candidate should have a good grounding in Python. Knowledge of random forest, data linkage and curation, and error propagation is desirable, but no essential.   

As this is project involves accessing NHS data, successful candidates will have to sign an honorary contract with RFL. 

This placement is now closed to applications.

Innovation Placement 2 with the NHS

In collaboration with the NHS, DiRAC is pleased to invite applications for a 6-month Innovation Placement focussing on The Application of AI in Cancer Detection Rates

Host organisation: Royal Free London (RFL) NHS Foundation Trust
Location: IIA operates a hybrid working policy, with up to two days a week office based, however, fully remote would be considered with regular, pre-defined F2F meetings. As per RFL policy, employees cannot work outside the UK.  

Background:

More people than ever before are affected by cancer; an estimated 50% of people will receive a cancer diagnosis within their lifetime. Detecting cancer at an early stage saves lives. Early diagnosis is improving thanks to new research that addresses the needs of specific groups as well as screening and early detection. However, access times are poor and UK patients are often being diagnosed when their cancer is at a later stage than our European counterparts, making successful treatment less likely.

Cancer referrals and treatments at Royal Free London (RFL) have increased significantly in the last 12 years. The RFL receives over 50% of all urgent cancer referrals in North Central London (NCL).

The RFL is already leveraging AI and automation to improve cancer care. We run one of 15 breast screening services supporting the MIA (mammography intelligent assessment (MIA) programme to test if AI can improve breast cancer detection rates, and RFL dermatology is testing AI to address health inequalities and improve early diagnosis of skin cancer. 

The Project

The aim of this project is to provide insight into the impact of MRI scanning on outcomes in the breast cancer pathway. MRI scan slots are a scarce resource that is in high demand; unnecessary use of the MRI scanners causes delay in the care of other patients which causes significant system pressures given elective care waiting lists are currently very long. In addition, MRI scans, whilst not resulting in exposure to ionising radiation, are unpleasant and have the potential to cause harm for patients if incidental findings are picking up that need further investigation e.g. with invasive biopsies. By evaluating use of MRI in the breast cancer pathway RFL seeks to understand:

• How are we currently using them? What types of patients (including their socio-economic status) and their clinical attributes are triggering an MRI scan, and how this is linked to a change in treatment plan, diagnosis and outcomes?
• How do patient outcomes differ where MRIs have been performed and where they have not?
• Is there unwarranted variation in the use of MRI?
• What would be the impact on the wider hospital system if changes were made to the use of MRI in breast cancer?
• How MRI scans are contributing toward changing health inequalities around access to care, and how we are affecting the patient experience and quality of life?

The project is designed to be accessible to candidates without healthcare data experience. Candidates will undertake training and shadowing sessions to understand how clinical data is generated and how errors can occur. The project will assess current practices, clean and organize data, and build a pipeline using data stewardship and data science. Statistical methods will be used to identify the target population and compare classical statistics approaches to machine learning. The project will also compare the impact of different data inputs and aims to produce work suitable for publication in peer-reviewed journals. The candidate will be expected to present their findings to both data and clinical groups across RFL, including:

• Cancer clinical practice group
• NCL cancer alliance
• Data steering group
  

Candidate profile

Candidates will be expected to apply the tools and techniques they have learnt or used during their studies from areas with an advanced approach to data science, into an applied healthcare project.

A suitable candidate should have a good grounding in Python. Knowledge of random forest, data linkage and curation, and error propagation is desirable, but no essential.   

As this is project involves accessing NHS data, successful candidates will have to sign an honorary contract with RFL. 

This placement is now closed to applications.

Innovation Placement 3 with the NHS/University of Liverpool

In collaboration with the University of Liverpool and their partnership with NHSE and HDRUK, DiRAC is pleased to invite applications for a 6-month Innovation Placement focussing on the link between healthcare data and medication data and outcomes.

Host organisation: University of Liverpool
Location: University of Liverpool or remote

Background:

Medicines are the most common intervention in human health. In the UK we spend £20 billion per annum on medicines, however it is not known if these medicines are used as intended or if they are having the effects they are intended to have. This is because health care outcome data is not routinely linked to medicines data. Moreover, medicines data is not all stored in the same place, rather is siloed across primary and secondary specialist care.  

To overcome this we are working with partners including but not limited to NHS England (NHSE), Health Data Research UK (HDRUK) to begin to address these limitations by identifying and linking medicines and health care data.  

This work includes a number of work streams:

• Identification of data sets and governance around their use 
• Permissions and transfer of data sets into a secure data environment 
• Curation of medicines data so they are available to be readily used for healthcare and research 
• Linkage of medicines to primary and secondary health outcome data 

We will then use the above to build tools for health care professionals and researchers to assess how medicines are being used across the health care system and with specific examples begin to show if they are being used as intended and if they are cost-effective and safe.  

The Project

The Project will model the effect of medicines based on the prescription histories of medicines and linked health care data. 

The project will use both ML and statistical models to assess how changes of uses of medicines occur through a patient period and patterns of common and rare medicines evolve. 

• This will require creation/adaption of dashboards, general and disease specific to determine how and where medicines are being used for both individual patients and groups of patients. 
• The prescription and patient health care data will need to be linked and then used to train a Deep Learning Neural Networks so that (i) heuristics can be generated and (ii) they can be retrained with each new data update. 
• Proto-heuristics on the following two areas; the effects of medicines on new patients, and poly-pharmacy on patients will be generated for both specific patients and groups of patients for further study. 
• The proto heuristics generated above will then be compared with the results of current statistics methods used in the above areas to determine their potential efficacy and to suggest improvements. 

These investigations will take place in a secure Health care data TRE and the data will be anonymised for the purposes of this project. 

Candidate profile

The successful candidate will posses the following experience:

• Good understanding of data and data structures and models 
• Good working knowledge of data bases 
• Coding skills in python and other languages 
• Statistical skills and use of relevant packages e.g R 

This placement is now closed to applications.

---

IBM Quantum Lab and Hartree Centre

In collaboration with IBM Quantum Lab and STFC’s Hartree Centre, DiRAC is pleased to invite applications for a 6-month Innovation Placement focussing on Quantum Machine Learning applications

DiRAC Innovation Placements are a great opportunity for doctoral students and early career researchers to work with technology leaders on a 6-month project that delivers research impact and develops key skills. Offering the potential to gain experience working collaboratively with the Hartree Centre and its Industrial partners, whilst fostering fundamental research competencies through mentoring, this competitive, fully funded internship will address current challenges with cutting edge solutions and provide useful insight to both the successful candidate and the mentoring partners.

You must get your supervisor or PI’s permission before applying for this placement. Participation in the placement scheme is allowed under UKRI’s rules, but only with your supervisor/PI’s consent. We will do our best to be flexible; part time working can be arranged as long as the placement does not exceed 9 months. The placement will be fully remote.

Background:

The Hartree Centre is within the Science and Technology Facilities Council (STFC) and part of UK Research and Innovation. They are funded by UK Government to support businesses and the public sector on their digital journey. The Hartree Centre forms an important part of the UK’s research and innovation ecosystem, alongside a global network of academic research communities and technology partners.

The new Hartree National Centre for Digital Innovation (HNCDI) is a collaborative programme between the Hartree Centre and IBM Quantum Network which supports UK businesses and the public sector to explore adoption of innovative new digital technologies such as AI and quantum computing.

The Hartree Centre also hosts quantum simulators developed by Atos which will be available to the project. The Atos Quantum Learning Machines allow researchers to develop and experiment with quantum software and can simulate quantum computers of up to 38 qubits.

Project Description:

The project provides the successful candidate with the unique opportunity to work with the Hartree Centre’s team of quantum software engineers and collaborators at IBM Quantum and Atos to develop new quantum machine learning applications to solve real-world problems and learn the skills to apply quantum computing methods to their own research field.

The successful candidate will work with the application of Quantum Machine Learning and Quantum Optimisation algorithms towards the solution of industry-relevant challenges in engineering, logistics, chemistry, drug discovery and materials science. Furthermore, the candidate will be responsible for or involved in the design of hybrid classical-quantum algorithms, which will be executed on classical computing facilities (HPC), quantum simulators (ATOS QLM) and real quantum hardware using the IBM Quantum services provided via the HNCDI hub.

Prerequisites:

Essential:

• Mathematically competent, with some experience of writing scientific software
• Good knowledge of quantum information theory
• Strong background in Machine Learning Techniques

Desirable:

• Experience of working within Qiskit, and opensource software libraries
• Experience of Python Programming
• Experience of QCompute

Personal attributes and qualities:

• Excellent communication Skills
• Motivated team player
• Practical and hands-on approach to problem solving and projects

This placement is now closed to applications.

---

Quantum Innovation Placement in Radio Astronomy

In collaboration with Google and the University of Manchester, DiRAC is pleased to invite applications for a 6-month Innovation Placement focussing on a Quantum Machine Learning pulsar classification model in the field of Radio Astronomy.

DiRAC Innovation Placements are a great opportunity for doctoral students and early career researchers to work on a 6-month industry-academic project that delivers research impact and develops key skills. Offering the potential to gain industrial experience by working collaboratively with businesses and fostering fundamental research competencies through academic mentoring, these competitive, fully funded internships address current challenges with cutting edge solutions. The projects are designed to tackle particular research challenges and provide useful insight to both the academic university partner and the industrial partner.

You must get your supervisor’s or PI’s permission before applying for this placement. Participation in the placement scheme is allowed under UKRI’s rules, but only with your supervisor/PI’s consent. We will do our best to be flexible; part time working can be arranged as long as the placement does not exceed 9 months. The placement will be remote.

Background:

Google is advancing state of the art quantum computing and developing tools for researchers to operate beyond classical capabilities. This project will utilise Google’s Qsim Quantum Simulator – a full wave function simulator written in C++ which uses gate fusion, AVS/FMA vectorised instructions and multi-threading using OpenMP to achieve simulations of quantum circuits. Depending on progress, the possibility exists for the project to ultimately transition to Google’s quantum processor hardware.

The project’s academic mentor is Professor Anna Scaife from the University of Manchester’s Astronomy and Astrophysics Theory Group. Professor Scaife is head of the Jodrell Bank Centre for Astrophysics Interferometry Centre of Excellence and a Fellow of the Alan Turing Institute, where her research focusses on AI discovery in data intensive astrophysics.

Project Description:

The project will work with the Quantum Machine Learning (QML) pulsar classification model from Kordzanganeh, Utting & Scaife (NeurIPS 2021) and will be divided into two phases:

Phase 1: The initial learning phase will be devoted to understanding the basics of Quantum Computing and the Google quantum simulator training materials. The successful candidate will also gain a familiarisation of the details of Kordzanganeh, Utting & Scaife (NeurIPS 2021) and compare and contextualise that model with respect to the Google materials. During this period the student will create an implementation of the QML pulsar model for deployment on the Google quantum simulator.

Phase 2: Using the Google quantum simulator, the student will evaluate the performance of the QML pulsar model. This will include benchmarking a selection of different quantum encodings and encoding depths, evaluating the performance impact of the entangling layers in the multi-qubit model, determining how the algorithm will scale for different volumes of train/test data, and undertaking a trade-off analysis exploring the computational cost of using different numbers of qubits vs. different levels of parallelisation as a function of machine learning model performance. Depending on progress in this phase of the project there is the possibility of using Google’s quantum hardware to compare the performance of real hardware to the quantum simulator.

Prerequisites:

This project is suitable for students who have experience in Python programming. Experience with deep-learning is desirable but not required as the principles of QML will be covered in the first 3 months of the internship. No previous experience of quantum computing is required. Code for the QML pulsar classification model is available in Python (using the pennylane QML library).

Placement:

The successful candidate will work with Professor Anna Scaife as a research point of contact, with support from collaborators in the University of Manchester Physics Theory Group. Access to Google materials and systems will be facilitated by Google High Performance Computing consultants and Google.edu, a service which assists students and researchers to use Google technology.

This placement is now closed to applications.

Designing and developing simulations to understand climate-driven migration, mentored by Save The Children’s Innovation Project Lead William Low and Brunel University London’s Reader Derek Groen.

Background

Climate-driven migration is a major challenge for the global community, and frequently cited as a main cause for future societal disruption. However, little is understood about the direct impact mechanisms of climate change on the livelihoods and coping mechanisms of affected people in the Global South. Simulation approaches have previously been successful in forecasting conflict-driven migration (https://www.nature.com/articles/s41598-017-13828-9), and provide a good environment to study the interplay between climate events and the migrations of affected communities.

Flee (flee.readthedocs.io) is a widely recognized agent-based migration model that represents people as individual agents that move between locations and make autonomous decisions based on user-specified criteria. The approach is unique in the migration modeling context in that it works with explicit assumptions, and that it can be used without the need to train it against (often ethically sensitive) humanitarian data. Our latest version has the added capability to explicitly support agents with different ages, ethnicities and economic circumstances, and Brunel University and Save The Children currently collaborate to incorporate support for disaster- and climate-driven migration. This placement will operate within that context, with a specific focus on a particular climate hotspot region.

The Team proposing this project consists of the Migration Modelling Group at Brunel University London (Dr. Derek Groen, Dr. Diana Suleimenova and Dr. Alireza Jahani) in collaboration with Save The Children International (Innovation Lead William Low and Data Scientist Auke Tas). The Migration Modelling Group has led the development of the Flee simulation code, while Save The Children has coordinated efforts to investigate the benefits of Flee to forecast migration in countries such as Burundi, Ethiopia and Nigeria.

Project Description

In this project we are interested in simulating how communities adapt and respond to climate-driven events such as flooding. The project is divided into two phases, each lasting approximately 3 months.

Phase 1: The initial learning phase is devoted to gaining essential technical knowledge, as well as a well-rounded understanding of the climate and community dynamics in a selected geographical climate hotspot region (e.g., Bangladesh, Somalia or Central America) as we know it today. Deliverables from this phase include an overview report of the migration and climate situation in this region, as well as a prototype simulation sketch which captures the essential communities, climate events, and coping/migration behaviors.

Phase 2: The second phase will focus on the refinement and implementation of the simulation sketch into a working Flee model, and extract preliminary findings by analyzing the behaviors of this model as a running simulation.

Prerequisites: This project requires prior experience in Python programming, and applicants will benefit from basic knowledge of graph theory and at least one type of model (e.g. Nbody, SPH, or any other).

Placement: The successful applicant will work in collaboration with the Migration Modelling Group at Brunel University London and Save The Children International in both phases. Dr. Derek Groen will take the main mentoring responsibility, but the applicant will be encouraged to work with other members of the team as they find convenient.

This placement is now closed to applications.