CryoDANN: Using AI to see proteins more clearly at Diamond Light Source
Researchers at the electron Bio-Imaging Centre (eBIC), based at Diamond Light Source, are using state of the art cryo-electron microscopy to study virus and protein structures in 3D at very high resolution. Seeing proteins in 3D can help researchers to understand their function, how they interact with biological molecules, and enable discoveries that could improve our health.
The Challenge
Researchers build 3D structure of proteins using the information from tens of thousands of electron microscopy images. They limit the number of electrons used to take each individual image to avoid damaging the sample, but because of this the resulting images are grainy and it’s hard to see the difference between useful information and background noise.
Computer software is needed to convert the 2D images into highly useful 3D images. Researchers currently support this process by improving the quality of the noisy electron microscopy images. They do this by manually selecting which information from the images is useful and which is background noise. This process is time consuming, requires significant computing resource and is prone to human error.
“The major bottleneck at eBIC is our ability to provide fast and reliable 3D reconstruction of an object, a process which normally takes weeks, if not months. CryoDANN has sped up this process, allowing researchers to achieve the same or better result quicker. We’ve also found that in many instances it has improved the quality of the data, allowing a high-quality 3D model to be constructed when this would previously not have been possible.”
Yuriy Chaban, Principal Electron Microscopist for the electron Bio-Imaging Centre (eBIC).
What did we do?
The Ada Lovelace Centre (ALC) started a joint project with researchers from Diamond and STFC Scientific Computing to see if AI could process the images to reduce noise and increase resolution. A new AI machine learning programme, called CryoDANN, was developed that can recognise which information in the electron microscopy images is useful and apply this learning to different applications. CryoDANN was also able to make decisions about the usefulness of information faster and at much finer detail than researchers doing the process manually.
“This AI approach to sorting and cleaning noisy images replaces a tricky manual task with a reliable automated solution. This reduces a major bottleneck for researchers at eBIC and delivers on the ALC mission to provide innovative computing for ambitious science”
Paul Quinn, Ada Lovelace Centre Director
What is the impact of this project?
CryoDANN allows for the creation of more accurate and detailed 3D images of proteins and viruses than was previously possible. This is allowing researchers to better understand how molecules and atoms fit together and interact, something which is essential for applications such as vaccine development and structure-based drug design. The software takes the human out of the loop, speeding up the time it takes to generate images and making the process more efficient and streamlined. The new method can also give researchers quick feedback on the quality of the data they are collecting, which allows them to make changes mid experiment if needed – saving time and resources. CryoDANN software is available for all researchers using the eBIC facility at Diamond.
"The microscopes at Diamond Light Source are in high demand. Our method gives users feedback quickly and correctly, allowing them to make the best use of their time. Our aim is to help researchers process their data as quickly and efficiently as possible, so they can get the high-resolution images they need to understand viruses, vaccines, drug discovery and the molecular basis of life,”
Tom Burnley, Molecular and Cellular Electron Microscopy Group Leader, Scientific Computing
