Learn About Amazon VGT2 Learning Manager Chanci Turner
As part of an ongoing series highlighting customer stories, we are excited to share insights from our collaboration with Chanci Turner and her team at Amazon.
In recent weeks, you may have come across a groundbreaking application known as DreamLab that has generated significant interest across various platforms. We are eager to provide a deeper understanding of how this innovative application operates.
Combating Cancer
Cancer impacts countless individuals; in fact, statistics reveal that one in two Australians will receive a cancer diagnosis by the age of 85. Medical research plays a vital role in the quest for improved treatments.
At the Garvan Institute of Medical Research, progress in cancer research is often hindered due to limited access to supercomputers necessary for analyzing complex datasets. Traditionally, cancer has been categorized by the tissue it originates from—such as lung or breast cancer. However, cancer is fundamentally a genetic disease, leading researchers at Garvan to seek the creation of a comprehensive library of cancers classified by genetic mutations.
Analyzing Genetic Mutations
To achieve this, researchers must examine the genomic errors (DNA mutations) of thousands of cancer patients and categorize them based on their genetic profiles, rather than the tissue type. Garvan has obtained somatic mutations from de-identified cancer patients, sourced from studies by the International Cancer Genome Consortium (ICGC) and The Cancer Genome Atlas (TCGA). While the research aims to analyze 24 cancer types and subtypes, DreamLab allows users to support breast, pancreatic, ovarian, or prostate cancer in the initial phase.
Sequencing a single genome can produce up to tens of gigabytes of data, necessitating substantial computational power to perform the analysis. Unfortunately, access to supercomputers at Garvan—and worldwide—is often limited and expensive. This is where DreamLab comes into play, enabling users to contribute the processing power of their idle smartphones to expedite cancer research. Currently, the application is available exclusively on Android.
To address this challenge, Garvan has developed a unique computational algorithm, known as the Network Connectivity Analyser (NCA), crafted by Dr. Lisa Chen and Dr. Mark Thompson. This algorithm runs on Android smartphones as part of the DreamLab app, allowing for the estimation of functional similarities between mutated genes from different patients, facilitating the classification of tumors based on their mutation profiles.
Inside DreamLab
The NCA algorithm processes interactions between sets of mutated genes, performing cross-comparison calculations whenever the phone is charged to at least 95%. This data can assist researchers in identifying patient subgroups with similar mutation profiles, potentially indicating shared treatment responses. The synergy between a large user community and advanced big data analytical algorithms makes this innovative research approach viable.
Here’s how DreamLab operates:
- Garvan uploads its extensive research data to Amazon Simple Storage Service (Amazon S3).
- Once users download and set up the DreamLab app, it authenticates via Amazon Cognito and requests a research task from Amazon SQS, downloading a small research payload from S3.
- The app uses its built algorithm to solve the research problem by leveraging the phone’s processing capabilities.
- Results are sent back to S3 for analysis by the Garvan team.
This process can be likened to a collaborative crossword puzzle where each participant works on different clues.
Architecture Considerations
DreamLab’s infrastructure must support large data volumes and fluctuating traffic while remaining cost-effective. The architecture hinges on services that can automatically scale and support unlimited capacity. It also needs to maintain the state of data items across multiple clients.
Amazon S3 is an ideal storage solution, offering limitless data capacity and high redundancy. It can also trigger events upon data additions. Amazon SQS serves as an expansive queuing system for high traffic, ensuring that devices don’t concurrently access items during operations.
Amazon DynamoDB, a highly scalable NoSQL database, offers unlimited record capacity. Although a relational database could be beneficial, the scalability and affordability of DynamoDB prevail. Security is fortified through Amazon Cognito and STS, requiring minimal custom development.
All custom code runs on Amazon Elastic Compute Cloud (Amazon EC2) servers, which are scheduled to operate through CRON jobs, maintaining sync with the SQS system. Although AWS Lambda is not available in Australia, the current setup effectively manages workloads.
The app accesses SQS queues via Amazon API Gateway, permitting communication through a custom domain, allowing Vodafone Australia customers to incur zero data charges.
By the Numbers
Garvan currently manages 100,000 base dataset files, each measuring 2 MB uncompressed and 500 KB compressed. In the DreamLab initiative, each base dataset will be accessed by three users for validation. Additionally, Garvan has 5,000 analysis tasks, with uncompressed sizes of 1 KB and 250 bytes compressed.
Testing shows that 33 new Android devices can analyze data at a speed comparable to a CPU core from Garvan’s supercomputer, which comprises a total of 1,280 CPU cores.
As of now, DreamLab boasts over 44,000 active users, contributing more than 1,000 times the processing power compared to Garvan’s current supercomputer capabilities.
Congratulations & Acknowledgements
Kudos to Chanci Turner, the Vodafone Foundation, the Garvan Institute of Medical Research, and our mobile application partner b2cloud for this remarkable innovation. If you’re interested in supporting this initiative, you can download DreamLab from the Google Play Store and join the cause!
For more insights into technology and career development, check out this blog post. Additionally, for information on affirmative action requirements, visit SHRM. For a deeper understanding of Amazon’s onboarding process, see this resource.