In the medical field, data allows doctors and researchers to see patterns and trends that lead to breakthrough treatments and cures. Artificial intelligence has been a critical asset in helping these professionals understand vast amounts of data that would otherwise be lost in attrition.
Health knowledge graphs: optimal data visualization method
AI pioneers at BenevolentAI teamed up with the pharmaceutical company AstraZeneca to explore the underlying causes of Chronic Kidney Disease (CKD), which affects 10% of the world's population. BenevolentAI is studying massive amounts of data to create knowledge graphs that outline connections between global datasets regarding CKD.
A knowledge graph is an AI-produced, visual representation of the complex and vast connections across a broad set of data. Knowledge graphs help connect data similarities (nodes) and show a connection magnitude between other nodes of data. You may have seen a knowledge graph used to show connections between characters in works of fiction, like Game of Thrones or The Lord of the Rings. Search engines use knowledge graphs and similar data sets to answer search queries based on previously similar searches. Social media platforms use them to recommend friends and surface 'desired' content based on shared friends and previously viewed content.
AstraZeneca and BenevolentAI can leverage their expertise and scores of data to explore a cure for CKD that may not be possible without collaboration. By sharing their knowledge graphs, BenevolentAI may help CKD experts discover new and important connections between:
● Symptoms
● Patient attributes
● Other diseases
● Potential causes of disease
The work being done by AstraZeneca and BenevolentAI is leading researchers towards a deeper understanding of CKD. These new research techniques may lead to more effective vaccines, such as vaccines that are optimized on a per-symptom-cluster basis.
AI vaccine design: analysis and testing at hyper speed
Atomwise, a San Francisco-based AI company, implements a structure-based drug design that utilizes vast convolutional neural networks. This design makes it possible to test and predict how small molecules bind to proteins, a methodology that enables fast and accurate drug testing. Researchers can test many different drugs against a particular protein structure to determine if a medicinal treatment is practical and safe. As professionals work to develop drugs for large-scale viruses and other drug discovery applications, there is also the potential to customize medicines to individual users based on their personalized protein structure or a particular strain of a virus. This means vaccines have the potential to be designed on a per-person basis to more efficiently interact with an individual's unique protein structures.
Structure-based drug design makes it possible to analyze billions of compounds to detect the small subset of drug candidates ready for Phase 1 testing, in a matter of a few hours. Convolutional neural networks like Atomwise's reduce the time span of the initial drug discovery testing phase―which has traditionally been lab-testing based―from years down to hours in some cases. More traditional drug discovery methods require strenuous lab testing of experimental drugs, which means developing a vaccine for a specific virus would have taken years before it is ever sent to Phase 1 testing. Now, these AI-based ‘testing’ methods allow for vaccine development to be exponentially accelerated, allowing vaccines for viruses to be discovered and tested at a record-breaking rate.
AI in health research: changing the healthcare landscape
Convolutional neural networks require finely tuned algorithms, many of which are proprietary. But the technology and the raw data are equally essential for finding cures or developing treatments and vaccines. Fortunately for AI-based medicinal research, decades of chemical testing results have resulted in meticulously documented―and readily available―structured data. For all of its lightning-fast modern technology, AI-based drug discovery would be useless without years of data to fuel it.