Understanding the Value of AI in Healthcare

Why biomedical knowledge and its computation is a critical part of Digital Healthcare‍

If there is one thing that the pandemic is bound to teach humanity, it is the will to go with the flow and not plan too ahead in life. That said, counterintuitively, it seems important to be prepared and planned when similar outbreaks arise in the future.

A healthcare startup from Canada, BlueDot, in its report, shared that the outbreak was fast spreading in China even before the World Health Organisation (WHO) had reported it. The startup’s outbreak risk software helps mitigate the exposure to infectious diseases using Artificial Intelligence. 

It goes to show that healthcare organisations, governments and public health officials would have to rely on these AI-based systems in the future to prevent similar outbreaks from reoccurring. 

Though the future of digital health pertains to a myriad of technologies put together, whether it is Artificial Intelligence (AI), advanced health sensors or genomics, or all of them put together, digital healthcare is very much technology-based only on paper. However, what really matters in digital healthcare is how much of a cultural transformation can digital health and healthcare go through.

The usage of digital healthcare devices spanning across smartphones, wearables, internet applications and social networks is a broad spectrum in itself. The idea around monitoring an individual’s health status and well-being, powered by data and patient information has now become a norm.

Though much has been said about using artificial intelligence in digital healthcare, the real-world applications of AI have been quite a buzz ever since its disruption in the industry.

AI in the frontline of redefining the health ecosystem

With so much attention around using AI in healthcare, there is only a small set of stakeholders that truly understand the real impact and application of it.

The clear macro-level problems in healthcare, some of which are decades-old and traditional, include

• Lack of access to affordable and quality healthcare
• Inadequate resources, skewed supply and demand system, and distribution
• Lack of standardisation, error-prone diagnosis, unevolved care management
• Rising costs, poor healthcare coverage, low penetration by insurance
  

Despite the challenges, the introduction of AI in Digital Healthcare is poised to democratise the holistic care ecosystem, and the hopes are high enough that it supports both proactive and reactive healthcare. 

AI has been playing a major role in supercharging the fields of drug discovery, by simulating a wide variety of models and experiments on computers. Simply putting it, AI could dress off as an engine sitting on top of enormous amounts of patient data, generating relevant insights to all stakeholders and churning out effective clinical decisions. 

With this, the involved peers would be able to be better at underwriting risks and customise the relevant offerings. This also allows pharma research labs to be much efficient at drug discovery and development.

• Some examples include bridging the gaps in acute radiology and the subsequent report annotation and generation through tele-solutions deep learning. 
• Using images in pathology, AI helps generate precision-driven diagnosis. 
• With something as simple as our everyday push notifications and digital nudges, AI has its hands all over treatment for chronic conditions like obesity, hypertension and diabetes
  

The focus areas for Accel are AI in diagnostics, drug discovery, precision medicine and digital therapeutics with some recent investments made in these categories. The importance of the applications of AI and the large industry spend are the key attributes to double click on these categories.

Making bio-medical knowledge computable

In biology, scientists have traditionally shared insights and knowledge through written papers in the long-form format. The whole objective around making sense of all this knowledge and literature is directly contributed by deep learning techniques. 

When deep learning is applied to text in automatic translation at companies like Google and Facebook, auto-encoding and self-supervised neural networks provide a wide range of meaningful insights. 

The possibilities that come with self-encoding neural networks, along with digitised clinical records and scientific literature makes biomedical knowledge very computable. Having said that, deriving meaningful insights for better drug discovery and drug development is only possible from making sense of all the data that is already present in the databases.

When it comes to how this knowledge is synthesised, text-based data and literature is triangulated at the lowest possible layers along with molecular level data from DNA, RNA and proteins, and clinical healthcare records. 

Murali Aravamudan, CEO and Founder of nference said, “The digitisation of biology and electronic medical care records, coupled with the explosion of scientific literature led me to identify that I could solve meaningful problems of human life. This is the century of biology, and we need to conquer the subject of it fully with all the scientific advancements coming out there.”

Nference.ai is a Cambridge-based company that partners with various biopharma companies to solve the numerous problems in drug discovery, clinical research, life cycle management, clinical operations and commercial strategy. The comprehensive software platform converts a vast array of biomedical knowledge into computable data, resulting in answers to complex questions around disease biology and treatment outcomes.

‍Why is it important?

When someone falls ill, it is naturally because of a certain organ not functioning in its most efficient manner. Taking cues from the evolution of humankind on the planet, the ancestors were able to perform several models of reverse engineering from sick patients and figure out which organ exactly was troublesome. Gradually, a whole range of Ayurvedic medicine was developed over time. 

• Instructions like move, push and pop on an Intel microprocessor’s opcode are also reverse-engineered to execute the right program to get the desired output.
• An equivalent parallel scenario is in a DNA molecule where the four base pairs Adenine (A), Cytosine (C), Guanine (G) and Thymine (T) are bonded together to form a code that constitutes amino acids and proteins to form up a human DNA structure.
• This fundamental level of tinkering with human genome sequencing to make sense of underlying molecular products has been a slow process in the last 50 years. 1966 was when the original genetic code was cracked.
• All these years, several reverse-engineering tools were built into computational DNA, to try and understand the human body.
  

Murali says, “With the help of Artificial Intelligence (AI) in human genome sequencing and digital healthcare, along with clinical healthcare records, a massive digital explosion is poised to occur in the global healthcare ecosystem. This resembles the Cambrian radiation explosion that happened 540 million years ago, but all for the greater good.”

The applications of AI in the future

Before the COVID-19 pandemic , it would have been unimaginable to detect the symptoms of a certain virus infection like this. When SARS first broke out, it took a few years to make sense of all the symptoms when a virus like this infects the human body. 

But during COVID-19, it took only a matter of a few weeks for healthcare organisations like nference and Mayo Clinic to take data from past virus outbreaks like Influenza, 2003 SARS, and detect common symptoms like lack of sense of smell and taste. 

There has not been any code for something like a sense of smell in the past since it was not a problem that needed solving. But today, doctors are able to recommend patients to quarantine themselves immediately after such symptoms arise. 

A range of new sensory digital devices would now add to the existing human phenotypes and that makes it possible to predict any possible future health conditions. Intervention and diagnosis becomes easier as part of preventive healthcare, thanks to the applications of AI in digital health. 

• When it comes to genome sequencing, the primary idea is to identify whether the target molecules in the DNA structure would modulate or not. 
• Clearly, the problem is within the exponential searching within the gene structures to narrow down millions of preclinical molecules into mere five molecules that’s worth advancing.
• In the next stage, when the pre-clinical molecules move on to animal-model testing, lead optimisation comes into picture which also has its own set of unknown variables.
• From animal-model testing to phases one, two, three of pharma testing, artificial intelligence (AI) plays a major role across every stage.
  

Murali explains, “The way I see it, there are no limitations as to how AI finds its applications in the lifecycle management in the pharmaceutical industry. As care becomes virtual with telemedicine, whether it is acute care or chronic care, the lifecycle shrinks down to a matter of a few years, which would be decades without using AI.”

Challenges while adopting AI in Healthcare

There are naturally a whole plethora of challenges while trying to adapt to a new technology within a system. However, healthcare entrepreneurs are highly optimistic of a certain massive disruption in the near future. Though it takes much longer than anticipated, what is underestimated is the scale it reaches when the trend catches up at the appropriate time.

• The idea of transparency and how explainable AI can get to the doctors ecosystem is going to be a challenge, to be able to convince them to move on from legacy systems. Startups partnering with pharmas and working collaboratively with physicians will support in building credibility, trust and validation of the technology. 
• The technological possibilities of AI should not be oversold without peeling down all the underlying layers of it. Various diverse sets of systems would have to be built in order to avoid the possible bias by the underlying probability distribution, but rely more on reasoning.
• These above mentioned factors would have to be working in tandem with the ecosystem physicians and medical practitioners. The idea is to get the ecosystem comfortable with the idea of machines helping augment the decision-making mechanism in healthcare, with as much of a seamless integration into existing workflows as possible.