Genomics England is pushing the boundaries of how clinicians diagnose and treat cancer patients. The next step in its journey to change the face of medicine? A first-of-its-kind machine learning modelling framework.
4 months
to build an innovative ML modelling framework
4 modes
of data used in the analysis
100,000+
Data points or genomes
In 2013, Genomics England was given a monumental task: sequence 100,000 genomes from around 85,000 NHS patients affected by rare disease or cancer.
The aim was to create a database that could change the face of genomic research and our understanding of these diseases. Recruitment was completed in 2018 – and the next stage of Genomics England’s mission began.
To combine their comprehensive genome database with other public data including X-rays, radiology scans, pathology slide images to allow researchers to predict a patient’s prognosis or their response to specific treatments.
The problem? Extracting and analysing image-based data like radiology scans and pathology slide images sources at scale.
Together, Genomics England and Faculty aimed to build a completely new machine learning (ML) modelling framework that could achieve this monumental task.
Together, we’re building the basis to prove that – with the right platform, the right environment, and the right data science expertise – continued exploration of multimodal machine learning models is justified.
And, most importantly, Genomics England now has a better understanding of the strategic and technical best practices that should shape their next steps – and a clear view of the path to success.
Results
Execution
Ambition
Case study: Genomics England
Genomics England advances understanding of cancer with first-of-its-kind ML model
Get in touch
Find out more about our work in life sciences.
Revolutionise our understanding of cancer survival and progression
Genomics England’s Cancer 2.0 initiative aims to build the technologies to combine clinical, genomics, pathology, and radiology data to build a clearer picture of cancer diagnosis and progression.
A first-of-its-kind machine learning model and database
Genomics England partnered with Faculty to achieve one ambitious goal.
The multimodal problem: Putting cutting-edge data science into practice
Everything in our biology – from the physical appearance of our organs on an X-ray to the makeup of our genomes – is closely interrelated.
Something in our genomes might change the way we interpret something we see on an X-ray, and something we see on an X-ray might explain the results of an MRI.
And health and life sciences organisations are increasingly looking for ways to build machine learning models that reflect that complexity.
A machine learning model can be used to analyse almost any “mode” of data – from visual or image data to audio data, numerical data, and language data.
The challenge lies in including more than one different “mode” of data in one analysis. We call this multimodal data analysis.
It’s a new and rapidly developing field of research, and data scientists are still trying to understand the technical feasibility, the workflows and tools required and the performance and research benefits of multimodal models.
The first steps towards a health and life sciences revolution
With this innovative ML modelling framework, Genomics England will improve its understanding of AI’s role in cancer survival research.
Finally, we needed to ensure all this support was based on the most up-to-date research and the best health and life sciences strategy. We partnered with Genomics England to interview experts working within the multimodal research space, exploring the strategic context of the multimodal programme and providing insights to support the delivery of the programme.
The strategy
04
We also helped Genomics England understand the technical requirements for the machine learning development environment.
Which tools would be the best fit for future users conducting the analysis
How to design and set up the architecture
How to integrate existing platforms into the new architecture
On top of the codebase, we reviewed Genomics England’s planned architecture for their research environment and answered key questions like:
The architecture
01
Genomics England would be breaking new ground with this model – so it needed to be built on some of the most advanced machine learning techniques. With computer vision, Genomics England’s model would be able to scan visual data like X-rays and histology slides, extracting information that would usually go unnoticed by human clinicians to predict a patient’s chance of survival.
This ability to analyse a medical image with as much – or more – accuracy as the human eye is expected to change the face of medical diagnosis, prognosis, and treatment in the next few years – and it’s now at Genomics England’s disposal.
Computer vision
02
To get Genomics England on this new path, we needed to start by honing in on one area of the proposed project: predicting cancer survival rates.
We used publicly available data to build Genomics England the codebase for an innovative model that analyses four different types of data including genomic, clinical, histopathology and radiology data.
Each of these data types contains information that can help clinicians predict the likelihood that a cancer patient will survive. Combined, they should be able to predict survival chances with unprecedented levels of accuracy.
The codebase
01
We provided Genomics England with support in four key areas.
Genomics England needed Faculty’s data science expertise to understand – in detail – what it would take to build such a model.
But answering that question wasn’t just a technical challenge – it was pushing the boundaries of how machine learning is used today.
Predicting better: Combining data for more accurate cancer survival rates
Revolutionise our understanding of cancer survival and progression
In 2013, Genomics England was given a monumental task: sequence 100,000 genomes from around 85,000 NHS patients affected by rare disease or cancer.
The aim was to create a database that could change the face of genomic research. Recruitment was completed in 2018 – and the next stage of Genomics England’s mission began.
Genomics England partnered with Faculty to achieve one ambitious goal.
A groundbreaking database.
A first-of-its-kind machine learning model
To combine their publically available genome database with other public data including X-rays, radiology scans, pathology slide images to allow researchers and front-line healthcare providers to predict a patient’s prognosis or their response to specific treatments.
The problem? Extracting and analysing image-based data like radiology scans and pathology slide images sources at scale.
Genomics England partnered with Faculty to build a completely new machine learning (ML) modelling framework that could achieve this monumental task.
Everything in our biology – from the physical appearance of our organs on an X-ray to the makeup of our genomes – is closely interrelated.
The multimodal problem: Putting cutting-edge data science into practice
Something in our genomes might change the way we interpret something we see on an X-ray, and something we see on an X-ray might explain the results of an MRI.
And health and life sciences organisations are increasingly looking for ways to build machine learning models that reflect that complexity.
A machine learning model can be used to analyse almost any “mode” of data – from visual or image data to audio data, numerical data, and language data.
The challenge lies in including more than one different “mode” of data in one analysis. We call this multimodal data analysis.
It’s a new and rapidly developing field of research, and data scientists are still trying to understand the technical feasibility and the performance and research benefits of multimodal models.
Breaking new ground with AI – and changing the face of cancer diagnosis
Genomics England needed Faculty’s data science expertise to understand – in detail – what it would take to build such a model.
But answering that question wasn’t just a technical challenge – it was pushing the boundaries of how machine learning is used today.
We provided Genomics England with support in four key areas:
The codebase
To get Genomics England on this new path, we needed to start by honing in on one area of the proposed project: predicting cancer survival rates.
We built Genomics England the codebase for a first-of-its-kind model that analyses four types of data including genomic, clinical, histopathology and radiology data.
Each of these data types contains information that can help clinicians predict the likelihood that a cancer patient will survive. Combined, they should be able to predict survival chances with unprecedented levels of accuracy.
Genomics England would be breaking new ground with this model – so it needed to be built on the most advanced machine learning techniques. With computer vision, Genomics England’s model would be able to scan visual data like X-rays and histology slides, extracting all of the information that a human clinician would use to predict a patient’s chance of survival.
This ability to analyse a medical image with as much – or more – accuracy as the human eye is expected to change the face of medical diagnosis, prognosis, and treatment in the next few years – and it’s now at Genomics England’s disposal.
Computer vision
02
On top of the codebase, we reviewed Genomics England’s planned architecture for their research environment and answered key questions like:
The architecture
03
We also helped Genomics England understand the technical requirements for the machine learning development environment and identified gaps in their current machine learning platform.
Finally, we needed to ensure all of this support was based on the most up-to-date AI theory and the best health and life sciences strategy. We partnered with Genomics England to interview experts working within the multimodal research space, exploring the strategic context of the multimodal programme and the next steps needed to deliver this programme.
The strategy
04
The first steps towards a health and life sciences revolution
With this innovative ML modelling framework, Genomics England has made a huge leap forward in its understanding of AI’s role in cancer survival research.
Together, we’ve proven that – with the right platform, the right environment, and the right data science expertise – multimodal machine learning models can be used for this purpose.
And, most importantly, Genomics England now has a full understanding of the strategic and technical best practices that should shape their next steps – and a clear view of the path to success.