Health researchers are showing success applying artificial intelligence techniques to identify and predict the spread of a variety of infectious diseases (IDs), as well as improve diagnoses, expedite vaccine development, and detect side effects.
Most notably, AI was used to detect the novel COVID-19 viral outbreak originating in Wuhan, China, days in advance of notifications from the World Health Organization and US Centers for Disease Control (CDC), by analyzing multiple data sources, including government documents and social media in over 100 languages. Such early identification could enable a more rapid and informed clinical response to ID outbreaks, reducing the spread and impact of such diseases.
We live in an increasingly interconnected world. Knowledge of the itineraries of billions of travelers enabled AI-based analytics techniques to accurately predict the spread of COVID-19. AI is showing great promise in fighting other IDs. In 2019, researchers at Flinders University in Australia used AI to completely design a new human influenza vaccine, thought to be the first of its kind.
The researchers developed two AI programs. One of them, SAM (Smart Algorithms for Medical Discovery), was trained on which compounds activate the human immune system and which ones don’t work in order to determine viable candidates. A second program generated trillions of different chemical compounds, which were ingested by SAM to look for matches.
Top candidates were produced and tested on human blood cells and animals. The resulting flu vaccine was found to be superior to prior vaccines in animals and is being tested in humans in a US clinical trial. This promising new AI technique promises to identify ID vaccines in a fraction of the time and cost of the conventional vaccine-development process.
“AI is a powerful tool for identifying a known or novel virus and predicting key features of the viral structure that may be strong targets for developing vaccines. There is also the challenge of analyzing a constant flow of new information about how a virus is causing disease. AI-based analytics allow you to rapidly find patterns in data and identify new or existing treatments quickly, ultimately saving the lives of patients.”
Rebecca Laborde, master principal scientist
Researchers are also using AI and machine learning to predict whether a vaccine will generate a sufficient human immune response and whether adverse events are likely. They can train such systems using pre- and post- vaccination “omics” datasets—such as genomes, proteomes, blood transcriptomes, metabolomics, cytokine profiling, and mass cytometry—to detect which foreign substances provoke an immune response and which cause adverse events. Additional data can help predict the effectiveness of the vaccine to produce antibodies in patients. Factors such as gender, route of administration (oral or injection), and pre-existing conditions have been found to impact vaccine responsiveness.
Which influenza strains go into a vaccine is decided once per year for each hemisphere. Biopharmaceutical companies have about six months to develop, test, license, and sell new vaccines. There isn’t time for large randomized clinical trials to test efficacy or long-term safety pre-launch.
As recently as 2010, two influenza vaccines caused serious adverse events, or SAEs: Pandemrix was found to cause narcolepsy in children ages 4 to 19, and Fluvax caused febrile convulsions in children younger than 5.
Given the potential large-scale health ramifications, there’s an urgent need to develop and use new technologies and approaches to efficiently and accurately identify safety issues.
“AI is starting to deliver on the potential of both speeding drug development and bringing safer drugs to market. These advances are being enabled by the availability of high-capacity cloud computing power combined with rapidly evolving AI tools and frameworks, available to the largest down to the smallest research organizations.”
Bruce Palsulich, VP of product strategy
The US FDA and CDC use the Vaccine Adverse Event Reporting System, which manages about 30,000 reports of AEs annually, mainly mild ones. The system uses AI and natural language processing (NLP) to analyze structured and unstructured datasets in order to identify a vaccine SAE more rapidly than with conventional approaches.
Researchers are also using AI-based neural networks and deep learning techniques to aid in the reading of images of organs to help them diagnose IDs more accurately.
Pneumonia is the largest vaccine-preventable cause of death in children under 5 years old. While medical professionals typically use chest radiographs to diagnose pneumonia, they can interpret those images differently based on factors such as clinical training, available resources, even fatigue and distraction. This rapid and accurate diagnosis expedites treatment when time is of the essence, or when the number of trained radiologists available is limited.
Key to using AI effectively in a clinical trial or healthcare setting is to take steps at the beginning to ensure system interoperability and incorporate diverse data types and large datasets into a single platform for analysis. Codifying data in a similar manner will help NLP and neural network techniques sort through text and images, respectively, in order to find relevant trends that inform more accurate decisions regarding drug efficacy and safety.
Learn more about Oracle Health Sciences Safety Cloud
Contact us to learn more about how we have help you manage your clinical data needs.