Radix and GSK optimise processes in vaccine development

Impact

With AI assistance, lab technicians can spend up to 6x less time on CFU counting and reporting. The solution achieves superhuman accuracy while meeting the rigorous standards of the biotech industry.

Executive Summary

Context

GSK, the leading vaccine company, is continually improving its vaccine development processes. These processes are long, strictly regulated, and held to the highest standards of safety and quality.

Goal of the project

Reduce vaccine development time by optimising processes with AI. GSK wants to ensure the accuracy and consistency of these processes in order to bring its vaccines to the people who need them.

Solution

An AI application that assists GSK’s lab technicians in a specific area of vaccine design: the counting and reporting of colony forming units (CFU).The solution optimises every step of the process and allows lab technicians to focus on other areas of their vaccine design work.

Case Study

The challenge

As a science-led global healthcare company, GSK strives to be the world’s most innovative, best performing and trusted healthcare company. Belgium is the heart of GSK Vaccines, with 9000 employees and three major vaccine sites, including the global headquarters of GSK Vaccines’ division. This represents the biggest industrial network of vaccines worldwide, as GSK produces two million doses every day distributed globally.

Technical Research & Development is an essential part of vaccine design. The department’s goal is to develop vaccines that will relieve the burden of disease. This requires a great amount of strictly regulated processes to ensure the highest standards of safety and quality. The processes involved are long, but crucial: new vaccines can take decades before they go to market.

GSK is substantially investing in AI, machine learning and data science to optimise vaccine design processes. Radix worked with GSK on one specific process of vaccine design: colony forming units (CFU) counting.

The briefing

One way to develop protein-based vaccines is to grow the proteins needed in the vaccine from the E. coli bacteria, which is the “workhorse” of the biotech industry. Throughout the growth process samples are taken and placed on Petri dishes. After some time, each bacterial cell should grow into a colony forming unit (CFUs). GSK’s lab technicians then analyse each Petri dish and manually count the number and type of CFU to determine the number of viable cells. This ensures that the growth process is proceeding correctly and is a critical step in vaccine production.

This process is tedious. The elements in the Petri dish come in many different shapes, and technicians need to identify and count every single one of them. Sometimes, CFU can overlap, making the task even more difficult. These constraints considerably slow down the process and make it very time-consuming for GSK’s technicians, who analyse 200 Petri dishes every week.

GSK wanted to help its lab technicians in this tedious task to optimise the process and vaccine development time, with two clear benefits: supporting the well-being of its technicians by allowing them to focus on other parts of their work, and reducing time to market.

The solution needed to be:

  • Human-centered and AI-assisted: humans are always kept in the loop during and after development, as those processes always need humans to ensure the highest degree of quality and safety.

  • Agile

  • Accurate and consistent

“We really wanted to focus on our technicians’ well-being by helping them with this tedious task. We knew right away that we could use AI and machine learning to optimise CFU counting. We developed a first internal proof of concept, which we wanted to take to the next level. That’s where Radix helped us. They developed a solution to gain time and consistency, while maintaining human-level accuracy – a precondition when dealing with something as vital as vaccines.”
Paul SmythSenior Manager Data Science Capabilities at GSK

The work

Radix started developing a solution that would use AI and machine learning to accurately and consistently count CFU.To achieve this, Radix advised GSK on the most effective approach to design the solution:

  1. Identify the business problem and the AI task that solves it.

  2. Navigate the map of AI and Machine Learning architectures that can solve this problem, and choose the most effective architecture for GSK’s specific AI challenge.

  3. Customise that architecture for the specific needs of GSK’s lab technicians.

In full collaboration with GSK’s team, the first working prototype of the solution was ready in 2 weeks. After 4 weeks, Radix had developed a fully-functioning internal web interface with back-end and front-end where technicians could upload pictures of CFU. Radix’s machine learning model would then count the CFU and identify all the elements therein. The application also allowed users to create a custom reporting.

GSK’s technicians and end users were involved from the inception to the completion of the project. They were able to give Radix direct feedback on this first prototype. The lab technicians then gave feedback every two weeks on the following sprints to adjust directions and priorities, improving the application every sprint.

Radix’s application saves time on each step of the process: from the upload of pictures to the CFU annotating, counting and reporting. GSK’s technicians can spend up to 6x less time counting CFU.

The end solution checks all of GSK’s boxes, and is:

  • Accurate (superhuman-level)

  • Consistent (the same image through the same pretrained algorithm will give the same prediction every time). Inconsistency in CFU counting may occur in vaccine development where technicians annotate and count things differently. The application made the CFU counting consistent, thanks to the model’s averaging effect: it learns from all users’ input.

“Our ambition is now to fully integrate the application in our CFU counting process. We were impressed with how quickly Radix was able to develop a working prototype. They made our team and our technicians an integral part of the project. That was essential for us. The application allows our technicians to save much needed time and to focus on other areas of their work in vaccine design.”
Paul SmythSenior Manager Data Science Capabilities at GSK
“Not only is it supremely exciting to be able to contribute to the acceleration of vaccine development through AI, but we are also very grateful to the team at GSK for the opportunity to collaborate with them on this project, and their openness and drive to apply the state-of-the art in AI to push the boundaries of what is possible.”
Laurent SorberCTO at Radix

The future

GSK is constantly trying to improve its vaccine development processes, to be able to bring their vaccines more quickly to the market and to the people who need them. The company is currently examining ways to expand Radix’s solution, for instance by expanding it to different types of elements or using it in different departments and processes (e.g. GSK’s production sites). Other use cases following the same structure are also in the works, as GSK is working with Radix on several other projects.

“We are very pleased to work with Radix throughout this journey. We really appreciate their experience, extended knowledge and integrity. They show great professionalism and made an effort to understand our business and the challenges we are facing. They constantly engaged with our internal teams and our lab scientists. At GSK, we pride ourselves on being at the forefront of data science and machine learning. Radix was able to seamlessly integrate with our team and deliver a result that ties in perfectly with our workflow.”
Paul SmythSenior Manager Data Science Capabilities at GSK