The pharmaceutical industry’s shift from big data to broad data and AI
From single tasks to enterprise-level challenges, life science companies are using artificial intelligence (AI) to help support their teams
The pharmaceutical industry has so much complex data, and the goal for many companies is to bring this disparate information together quickly and easily to drive insights, such as how to access patients who would benefit from treatments. Successful implementation of life sciences technology, such as advanced analytics and AI, can potentially be key to this transformation.
We see AI adoption continuing to increase in the pharmaceutical industry, particularly in the area of AI we refer to as “Narrow AI.” Narrow AI takes an AI system and trains it on a narrow set of data – lots of similar data – to do something extremely fast. This category of AI tackles a single task in a single domain. This approach is very effective with routine tasks based on common knowledge.
In my experience, pharmaceutical companies are increasingly using narrow AI in tasks, such as using it to aid in medical coding. It can be very effective at a targeted task with a slice of data, for example, IBM Watson Health has worked with Roche to predict the early risk of kidney failure for diabetics.1 The AI-based algorithm we built can predict if a diabetic patient will have chronic kidney disease within three years with 79% accuracy.1
While Narrow AI is very effective at these types of tasks, enterprises today need AI that can scale up. The algorithm to predict kidney disease cannot be applied to predicting the likelihood of something else, such as a mental illness episode. It’s effective at its task, but not scalable to different problems.
This is where “Broad AI” comes into play. Broad AI can integrate multiple data types to accomplish an enterprise-level business process. In healthcare, that means it might incorporate not just clinical data, but also include genomic, phenotypic, financial, scientific literature and imaging data. Broad AI is the ability to learn many tasks in a domain. It brings together learning and reasoning enabling training with fewer examples across more tasks. With Broad AI a change in data does not require a change to the model, because the system adapts itself.
Broad AI can deliver faster time to market and value compared to Narrow AI. Most importantly it provides the mechanisms to scale and be economical.
The pharmaceutical industry, as well as other industries, is very early in its journey to Broad AI. But as we are no longer in a Big Data world, but rather a Broad Data world, I believe leaders recognize the value of harnessing a myriad data types to help drive more insights and decision making.
Here’s an example of Broad AI from IBM Research that applies to the pharmaceutical industry and other industries. IBM RXN for Chemistry is an AI system that helps predict organic chemical reactions. Users can create projects and collaborate on complex, multi-step reaction synthesis or on novel chemical reaction designs. The research team developed a language translation machine learning system and fed it with examples of chemical reactions – not rules or training from chemistry to build its predictive model.
Based on my experience, I believe Broad AI could potentially help clinicians and researchers:
- Improve the precision of predictions by combining data from the EHR with genomic data to develop patient cohorts. It could also contribute to removing bias and improving accuracy.
- Traverse into disease management, by helping individuals leveraging more personal data via monitoring devices so they can manage their health.
- Become more prescriptive, in other words advising on possible outcomes no longer just predicting.
- Scale the applicability of predictions to answer many questions for many diseases relevant for an individual or cohort.
- Predicting the early risk of chronic kidney disease in patients with diabetes using real-world data. Ravizza et al. Nature Medicine