Allcyte: Breaking new ground in cancer precision medicine with AI and Google Cloud

About Allcyte

Founded in 2017, Allcyte is a biotech company that combines advanced microscopy with AI to help pharmaceutical companies develop new drugs and bring functional precision medicine to cancer patients.

Industries: Healthcare
Location: Austria

Tell us your challenge. We're here to help.

Contact us

About CLOUDPILOTS

A Google Cloud partner since 2011, CLOUDPILOTS enables digital transformation and cloud-based collaboration from its offices in Cologne, Vienna, and Zurich.

Allcyte scales its advanced microscopy solution to support better cancer treatment development for both individual patients and comprehensive translational research and development programs.

Google Cloud Results

  • Scales analyses to meet unique needs of both pharma research and precision medicine applications for individual patients
  • Reduces long-term storage costs 10x with downgrading options available on Cloud Storage
  • Automatically generates precise, reliable logs to trace every stage of the experimentation process

Analyzes TBs of microscopy data in days, not weeks

Scientists developing new anticancer drugs have to contend with a troubling statistic: as many as 95% of drugs that enter clinical trials do not reach the market. Even those drugs that are approved for routine use very often only help a subset of treated patients. One of the key reasons for this, the Allcyte team believes, is that conventional methods used by scientists to study anticancer drugs in the laboratory and by physicians to select treatments for individual patients in the clinic do not recapitulate the complexity of the human disease and can thus incorrectly suggest clinical effectiveness where there really is none. Allcyte’s mission, leveraging recent advances in microscope technology and deep learning, is to enable pharmaceutical companies to choose the most promising drugs to advance to clinical studies, and physicians to choose the best possible treatments for each of their individual patients.

Using high-throughput microscopy and image analysis supported by machine learning, Allcyte can directly examine the effect of potential new or existing treatments in fresh tissue samples from cancer patients, analyzing each single cell for its reaction to the drug, at unprecedented scale. Importantly, the team, together with scientists from the the Center for Molecular Medicine of the Austrian Academy of Sciences and Medical University of Vienna, has shown for the first time that, using this unique approach, treating patients with drugs previously found to be active in their own cancer tissue increases their chances of response (Snijder et al 2017 Lancet Hematology).

“Until recently, obtaining this level of resolution in a physiologically relevant model like patient samples was simply not possible,” says Gregory Vladimer, Chief Scientific Officer at Allcyte. “Previously, it was a case of spending weeks growing cells from cancerous tissue, then applying treatments to them in isolation. With our technology, we can determine whether or not specific drugs are going to work for an individual patient, leveraging all cells within the sample. And with cloud infrastructure, we can do it quickly. This precision medicine approach is unique to Allcyte.”

“Without up-front investment, we can access immense capacity on demand, for short periods. This flexibility remains central to our translational research. Collaborators often ask us about our platform’s scalability, and our Google Cloud-based solutions enable us to match whatever the job requires.”

Gregory Vladimer, Chief Scientific Officer, Allcyte

For scientists developing new treatments, Allcyte adds a new level of sophistication to the testing process by examining the effects of drugs on cancer cells in their primary context: surrounded by healthy cells and all of the complexity of their original microenvironment. The secret is scale: at capacity, the current infrastructure of high-throughput microscopes produce 250,000 images every 24 hours, amounting to 300 GB of raw image data. Powerful machine learning and neural network algorithms then analyze those images to identify 200 million individual cells and generate a detailed analysis of the drug response of a tissue sample.

Initially established in the blood cancer space, Allcyte is now moving into analyzing more complex solid tumor samples. “The samples for solid tumors are much more challenging to analyze,” says Robert Sehlke, Lead Bioinformatics Researcher at Allcyte. Using Google Cloud, the team has set out to adapt its deep learning solutions to fully characterize solid tumors.

“From the beginning, it was a huge advantage to be on Google Cloud and run with this ‘healthcare in the cloud’ idea,” says Gregory. “Without up-front investment, we can access immense capacity on demand, for short periods. This flexibility remains central to our translational research. Collaborators often ask us about our platform’s scalability, and our Google Cloud-based solutions enable us to match whatever the job requires.”

Using AI and deep learning to develop more complex analyses

In academic laboratories, computational resources are often shared between multiple projects. For novel and young biotech companies, however, the up-front cost of hardware capable of running large calculations is often prohibitive. On-demand, scalable cloud solutions are therefore an attractive, even necessary option. For a typical analysis, Allcyte divides a patient sample into a plate containing 384 small petri dishes, each loaded with individual drugs, to record, image, and capture a high quantity of singular drug responses.

“First, we want to know how the sample behaves with no intervention,” says Robert. “We observe how cancer cells respond to the drug, but also how the cells around it from healthy tissue react to the treatment. This is done by taking images of each of these petri dishes. That’s where the cloud comes in: with image analysis, we need to pin down each and every cell and its location, count them, and quantify how they interact with, for example, immune cells. We do that through machine learning, and Google Cloud provides a flexible way of delivering the computational power needed to analyze these images.”

Allcyte is embracing AI as it extends into the analysis of more complicated samples taken from solid tumors. “To address the sheer variety of different cell morphologies, we needed to branch out to convolutional neural network (CNN) architectures, a specific type of artificial neural network used in image recognition,” explains Robert. “We need GPUs for that kind of analysis, and flexibility is essential. Without the cloud, it would have been a huge effort to switch from an infrastructure focused on processing simpler images towards a more deep-learning-oriented infrastructure.”

Scaling up from individual patient treatments to mass screening

Because the computational needs for each job vary greatly, Allcyte creates the capacity it needs for each task using Compute EngineGoogle Compute Engine. “We use Google Compute Engine instances for almost everything, as it provides the flexibility we need for our workstations,” says Gregory. “Even when we’re working to a smaller scale than the analysis of thousands and thousands of images, there’s a lot of analysis to be done. If we run into a dataset where 16 gigabytes of RAM is not enough, we simply take the development machine offline for a moment, scale it up to 64 or 100 GB of RAM, and get back online. It’s ready to use immediately.”

To screen samples at scale, Allcyte uses Google Kubernetes Engine, working with cloud consultancy CLOUDPILOTS to implement the more powerful GPU-based solutions needed to analyze complicated samples. “CLOUDPILOTS helps us use advanced tools like Cloud Build to take our software from code on GitHub and deploy it as a cluster on Google Kubernetes Engine,” says Robert. “We can write a new version of our software and commit it, and then Cloud Build takes care of all of the configuration details for us. Within an hour, we can go from approving a version of the code to actually being able to use it at scale.”

“In healthcare, every file needs to be traced extremely precisely. Google Cloud’s operations suite takes care of logging everything on our main image analysis pipeline, while features on Cloud Storage help us trace the history of every file. That reliable, precise tracing is invaluable for handling experimental data at scale.”

Robert Sehlke, Lead Bioinformatics Researcher, Allcyte

Documenting every stage with Google Cloud and Google Workspace

Extensive and precise documentation is vital at every stage of the process, from the development of algorithms to tracing images of samples. As Google Workspace users, the Allcyte team sees useful synergies between their administrative tools, such as Docs and Sheets, and the architecture it has built on Google Cloud. “We can keep vital metadata information that we need for understanding experiments on shared team drives in Drive, and then access the same data from instances on Compute Engine,” says Robert.

“In healthcare, every file needs to be traced extremely precisely,” says Robert. “Google Cloud’s operations suite takes care of logging everything on our main image analysis pipeline, while features on Cloud Storage help us trace the history of every file. That reliable, precise tracing is invaluable for handling experimental data at scale.”

Optimizing data storage transparency and security

“For any biotech startup, especially one so interconnected with the clinic, data security is extremely important,” says Gregory. “When we made the decision to go with Google Cloud, it was the transparency behind data security that clinched it. The whitepapers from Google Cloud regarding data security, data processing, data ownership, and workflows have been incredibly insightful.”

Allcyte uses options in Cloud Storage to store the detailed data generated through its work according to its high security specifications, with the benefits of automatic backups and redundancy.

“Cloud Storage buckets are the primary and final destination for all of the raw data we generate,” says Robert. “Initially, it is available quickly and at low cost to applications that process it directly in the cloud. But we also use life cycle rules to downgrade items that have not been touched for, say, six months, into a cheaper tier optimized for long-term storage. That reduces our storage costs tenfold, in the long run.”

“Cancer therapies have matured from drugs that kill all cells, to drugs that kill some cells, to the possibility of reengineering a patient’s own immune cells as treatment. With Google Cloud, we don’t worry about infrastructure, meaning that as these ideas mature, we’re free to engineer our primary models, too.”

Gregory Vladimer, Chief Scientific Officer, Allcyte

Engineering new models in a fast-moving field

In collaboration with partners in the pharmaceutical industry, Allcyte has run successful translational research programs on a number of anticancer medications, to test their effectiveness ahead of clinical trials and also prior to deeper investigations. “We can even see the specific mutations that make patients more likely to respond to certain drugs, and predict the effects of those drugs in our model system,” says Gregory. “That has helped to confirm that we have a very robust model for these more complicated samples.”

Now, Allcyte is looking to use Cloud Life Sciences API to implement its solution. “With Google Cloud Life Sciences, we don't have to get our hands dirty with many of the more nitty gritty implementation details,” says Robert. “It abstracts a number of things away, scaling workflows with little additional work. That's an enormous timesaver both in the development stage, and then in deployment.”

By tweaking the architecture it developed on Google Cloud, Allcyte has gone from analyzing blood cancer samples, to more complicated solid tumor tissues. Now it plans to introduce a new dimension to analysis, using its confocal microscopes to record samples in 3D and analyze them with neural networks: the latest development in a fast-moving field.

“Cancer therapies have matured from drugs that kill all cells, to drugs that kill some cells, to the possibility of reengineering a patient’s own immune cells as treatment,” says Gregory. “With Google Cloud, we don’t worry about infrastructure, meaning that as these ideas mature, we’re free to engineer our primary models, too.”

Tell us your challenge. We're here to help.

Contact us

About Allcyte

Founded in 2017, Allcyte is a biotech company that combines advanced microscopy with AI to help pharmaceutical companies develop new drugs and bring functional precision medicine to cancer patients.

Industries: Healthcare
Location: Austria

About CLOUDPILOTS

A Google Cloud partner since 2011, CLOUDPILOTS enables digital transformation and cloud-based collaboration from its offices in Cologne, Vienna, and Zurich.