Listen to this article
By Pankaj Mondal, DrugPatentWatch writer
Machine learning, the most fundamental form of artificial intelligence, has started infiltrating the medical field, and it seems machines can play a crucial role in improving our health. A study of over 50 executives in the healtcare sector by TechEmergence revealed that by 2025 AI will be adopted on a broader scale.
If there’s one thing the healthcare industry has in abundance, it’s undoubtedly data. And machine learning algorithms work better if they are exposed to more data. The savings would also be huge. McKinsey reports that machine learning could save pharma and medicine around $100 billion annually because of greater efficiencies in clinical trials, better insights for decision-making and innovative tools that can help consumers, physicians, regulators and insurers make informed decisions.
Its ability to spot patterns in massive volumes of data gives machine learning a wide range of applications, some of which are discussed below.
- Improving Medical Diagnosis
The idea behind machine learning in pharma is not to replace the doctor but to enhance his medical expertise. Artificial intelligence programs take the entire knowledge that a physician has, which is everything he learned in medical school and while training besides his experience in treating patients, and scales it to unprecedented levels.
With the magnanimous amount of data available to doctors, from information related to new drugs to disease symptoms, drug interaction and how different patients treated in the same manner can have unique outcomes-the dexterity to gain access to this information and digest it is rapidly becoming a vital skill. And machine learning enables them to learn from that data and put it into good practice.
For example, Modernizing Medicine, a program that collates data from 3,700 providers and 14 million patient visits, is easily able to diagnose a rare condition, scroll through the available treatment options, and write a prescription within seconds. This saves time, which translates to greater efficiency and lowered costs. It may look impressive but future capabilities will undermine this feat. IBM recently acquired Merge Health Care, a firm with a repository of more than 30 billion unique medical images, which will train WATSON to diagnose diseases.
Machine learning can also prohibit recidivism by assisting follow cases and make additional recommendations. At St. Jude’s Medical Center and Vanderbilt University Medical Center, electronic medical records are attached to AI. The doctor using them sporadically gets a pop-up, elucidating on how specific genetic traits can impact the patient’s condition or how a new medicine could improve their health. By clicking the pop-up, a physician can have a better idea of the disease and prescribe the most effective treatment. These electronic records aren’t only saving space and time, but also actively helping doctors recommend better treatments and making them aware of the nuances before them.
Radiologists in China, which is home to the highest number of lung cancer patients in the world, have started using AI programs to improve medical diagnosis in reading x-rays and CT scans and identify suspicious nodules and lesions in lung cancer patients.
- Better Patient Care
Physicians often have a difficult time keeping track of signals from test results, charts, and other metrics. The task of integrating and monitoring all of these data for a huge number of patients while making real-time decisions is also equally difficult.
Recently a team from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) created a program called “ICU Intervene” that takes massive amounts of intensive-care-unit (ICU) data and finds possible treatments for various symptoms. The AI system makes real-time predictions with the knowledge from past ICU cases to make recommendations for critical care besides explaining the reasons behind the decisions. The main objective is to use data from medical records to predict actionable interventions and improve healthcare. This machine learning program could help physicians in the ICU, which is a high-demand, high-stress environment.
Previous works in clinical decision making has largely focused on outcomes like the likelihood of death while this method predicts actionable treatment. Furthermore, the system uses a single model to predict multiple outcomes.
More importantly, ICU Intervene predicts far into the future. This model, for instance, can predict whether a patient will require a ventilator five hours later rather than only 30 minutes or an hour later.
The researchers will enhance ICU Intervene in the future to provide greater individualized care and give improved reasoning for decisions.
- Predicting The Next Epidemic
Epidemics are generally caused by numerous factors like a change in the ecology of the host population, introduction of an emerging pathogen to a host population or a genetic change in the pathogen reservoir. An epidemic moves with the speed of light, infecting a large number of people within a short time, usually two weeks or even less. In some cases, it can spread to other countries and affect a substantial number of people.
Because of the alarming risks associated to epidemics, it has been the highest priority to learn more about them, the way they spread and from which species. And now Malaysian startup, Artificial Intelligence and Medical Epidemiology (AIME), has managed to give users the exact location and date of the next dengue outbreak, 3 months ahead. The program also recommends anti-dengue measures for the infected area within a 400-meter radius.
With more than one third of the global population living in areas that are prone to dengue infection, it has becoming a leading cause of death in many countries. Although it rarely affects the US, it’s endemic in Puerto Rico, Latin America, the Pacific Islands and Southeast Asia.
The system from AIME analyzes public health data along with data from other sources like wind speed, weather, a location’s proximity to massive water bodies, previous outbreaks and anything that influences the behavior of mosquitoes responsible for carrying the disease. It also looks into other factors like population density in a given area, income level of people and their health records.
It’s not dengue alone that can be predicted by today’s AI. Scientists from the University of California, the University of Georgia and Massey University have created an AI model that can possibly pinpoint potential hubs for filovirus infection with 87% accuracy. Filoviruses usually infect bats but can be transmitted to humans and have the potential to inflict devastating consequences-Ebola, for example.
Instead of predicting the next Ebola outbreak, the model analyzes the last spillover event and makes predictions based on the elemental traits of filovirus-positive bat species, says David Hayman from Massey University.
- Helping Farmers Fight Famine
Approximately 800 million people in the world depend upon cassava roots as their basic source of carbohydrate. Similar to yam, the starchy vegetable is eaten like potatoes, but it can also be ground into flour for preparing cakes and bread. It’s the sixth most produced heavily-consumed plant in the world mainly because of its ability to grow in places where other crops don’t. Sadly, it’s also vulnerable to pests and diseases and can wreak havoc on huge fields of the vegetable.
Recently, researchers from Makere University in Uganda teamed up with experts to start the Mcrops Project in their quest to develop a machine learning system that can fight cassava diseases. Using cheap Smartphones, farmers can take pictures of their plants and use computer vision that can identify signs of the four important diseases responsible for demolishing cassava crop. So farmers know if they need to rip their crops or spray them. The system has been able to detect cassava diseases with 88% accuracy.
Mcrops also uses the images to find out patterns in disease outbreaks, which allows officials to halt epidemics that can possibly lead to famine. The team is now planning to use the technology to study banana diseases and automate identification of crop pests.
- Combating Cancer
About 14 million people are diagnosed with some form of cancer while 8.8 million people die of the dreaded disease every year. Detecting cancer at an early stage can dramatically improve a patient’s survival rate besides reducing the danger of the disease recurring. The best way to spot cancer early is through screening.
DeepMind, owned by Alphabet and IBM have been using their technology to fight this threat. DeepMind has teamed up with renowned clinical partners to train its AI to plan cancer treatments by identifying regions of healthy tissue from tumors in neck and head scans.
IBM’s Watson can analyze images and evaluate patient notes to identify tumors accurately in up to 96% of cases. Currently, the system is being used to help diagnose lung, breast, cervical, colorectal, prostrate, gastric and ovarian cancers.
At the University of Texas MD Anderson Cancer Center, the AI program is sifts through huge genetic data generated by patients’ cancers and directs physicians to the treatments that will give patients the best chance of improving their lifespan.
- Drug Efficacy Detection
The success of personalized drugs largely depends on the ability to identify patient sub-populations, which can be facilitated with precise diagnostic tests based on biomarkers. With the huge amount of proetomics, metabolomics or genomics data, identifying the most effective biomarker is a complicated task. Huge amounts of patient Omics data are being accumulated. Unfortunately, there are no tools in place to extract the required information from the data.
But with many machine learning companies rewriting the code for drug discovery, the implications will possibly be far ranging in the years to come. Benevolent AI, a leading AI company in the UK, is one leader that uses machine learning for discovery of drugs and diagnosis of diseases. Their system recently successfully identified biomarkers in Amyotrophic Lateral Sclerosis (ALS) also known as Motor Neuron Disease (MND). The technology was used to review billions of sentences from countless scientific research papers and abstracts. Following this, they started finding direct relationships between the data and transformed them into “known facts.” These known facts were then curated by the system to develop many hypotheses against qualified criteria. The scientific team then evaluated the validity of the hypotheses and picked 20 triaged biomarkers that were worthy of being further explored. The company later whittled down to five compounds, which were tested on ALS patient cells.
The data might show that a particular protein upregulates a given gene which may not be directly related, propelling researchers to find drugs in an entirely different arena, says Ken Mulvany, Chairman of BenevolentAI. Consequently, this model is able to find novel targets with data mining.
- Facilitating Clinical Trial Process
The clinical trial process is the most dangerous and cumbersome part of drug discovery. The fact is that nothing is predictable or certain in a clinical trial. Clinicians face numerous challenges while conducting a clinical trial like recruiting and retaining patients, spiraling costs, complexity of the trials and many regulations.
The invention of Internet of Things (IoT), mobile and wearables have allowed people to convey important information effortlessly. This offers a way to capture relevant data from patients in a convenient and continuous way. Patients can now easily share their information for clinical trials with the touch of a button. Also, the data captured is precise, contextual and of high quality.
With the improvement of data capturing technologies, the opportunity to leverage qualified data into an AI program will reduce patient risk and improve pharma’s quality and time. Boston-based biopharma company, Berg Health, has started using its unique AI-based platform, Interrogative Biology, which enables them to identify biomarkers for drug discovery and keep a track on patient responses during clinical trials. They build models using the platform with the patient’s own biology and monitor patient response on a biological level.
Another example of machine learning in clinical trials is the ATACH-II app that provides assistance in assessing patient eligibility, pre-screening and randomization.
- Revolutionize Pharmaceutical R&D
Bringing a new drug to the market costs over $1 billion in R&D expenses and takes about 12 years. Industry leaders are now considering implementing effective methods of approaching this process and machine learning seems to be a potential solution.
The world’s leading drug companies are turning to machine learning to improve the hit and miss business of finding new drugs. GlaxoSmithKline unveiled a $43 million deal in the arena in 2017. Other pharma giants like Sanofi, Johnson & Johnson and Merck & Co are also exploring the potential of big data to aid streamline the drug discovery process.
The concept is that machines, which are adroit at pattern recognition, can scrutinize vast amounts of new and existing metabolic, genetic and clinical information to decipher the complex biological networks that rationalize the occurrences of diseases. This, in turn, can help in the identification of drugs likely to work in given patient populations, while steering companies away from drugs that are likely to fail.
Machine learning is still in its infancy and will not be able to replace a doctor. But its ability to understand natural language like clinical notes and structured data like numbers and dates is being seen as the fourth industrial revolution, for which the pharmaceutical and healthcare industries will be the biggest beneficiaries.Copyright © DrugPatentWatch. Originally published at 8 Applications of Machine Learning in The Pharmaceutical Industry