5 top use cases of AI in the pharma sector
You will notice that it’s common for pharma companies to team up with tech innovators to successfully deploy AI. Accenture conducted a survey where 61% of the respondents reported at least a 5% increase in profit after partnering with a tech vendor, with 76% of pharma executives citing effective partnership as a key success factor.
Here are 5 top applications of artificial intelligence in pharmaceutics.
The Congressional Budget Office reports that the R&D costs of developing a new drug can exceed $2 billion, which includes research and clinical trials.
Deploying AI in pharma enables researchers to sift through enormous datasets, such as small molecule libraries and spot disease patterns, and learn which chemical compositions can be a good fit for various biological targets. AI can generate chemical compounds either as a text string or as a graph architecture. It’s important to validate the resulting compounds, as many of them will not make sense, could be toxic, or could contain a component that shouldn’t be a part of any drug.
In addition to discovering candidate compositions, scientists can use AI algorithms to parse medical literature on how to best synthesize the drug and design clinical trials. Research shows that pharmaceutical artificial intelligence can cut drug synthesizing and screening time by 50%, saving the pharma sector up to $26 billion in annual expenses.
There are many great examples of pharma companies deploying AI solutions to facilitate drug discovery. For instance, GSK, a British pharmaceutical company headquartered in London, partnered with California’s Vir Biotechnology during the pandemic to accelerate COVID-19 antibody discovery with the help of AI and a human gene editing tool, CRISPR. Vir already had an antibody platform that it deployed to discover drugs for different respiratory pathogens in the past. And now, in this collaboration, they discovered sotrovimab, an antibody that binds to a SARS-CoV-2 epitope to neutralize COVID-19.
In another example of collaboration between Europe and the US, a French pharma and healthcare company Sanofi partnered with California-based biotech innovator Atomwise to discover and synthesize drug compounds for five different targets. Sanofi wanted to steer clear of the traditional drug discovery approach and paid Atomwise $20 million upfront for their innovation and AI capabilities.
You can find more information on AI in drug discovery on our blog.
AI has many applications in clinical trials. One of them is identifying the right candidate participants. The technology can analyze patient data, genetic information, doctor notes, and other information, and pick people who are eligible for a particular trial. AI can even help decide on the optimal population size based on the existing description of similar trials.
86% of clinical trials fail to recruit enough patients within their target time frame. One-third of phase Ⅲ clinical trials have to stop due to recruitment-associated challenges.
For instance, IBM Watson relies on analytics and natural language processing (NLP) to analyze patient information. The tool can handle unstructured data, like doctor’s notes, and produce an insightful patient summary. Clinical researchers use these highlights to select and recruit patients.
As AI helps pharma companies to find patients, it also works the other way around. Antidote, a clinical trial patient recruitment platform, uses NLP to analyze their text and screen them for trial inclusion/exclusion criteria. It requires patients to answer a few simple questions on its platform and suggests a list of trials that the person can join.
Deploying AI in the pharmaceutical industry offers multiple opportunities to improve the drug production process. The technology could:
Assist in drug quality control. AI can inspect drugs on the conveyor belt and spot defects, such as damaged packaging. Moreover, the technology can identify any potential issues by analyzing manufacturing data, like quality control tests. For instance, AstraZeneca employs machine learning to analyze drug images looking for defects, while Merck applies AI to spot problems in vaccine vials.
Facilitate predictive maintenance. AI can monitor equipment on the production line and identify potential defects through sensors that measure equipment vibration, temperature, sound, etc. This gives employees time to fix the device before it breaks down, halting production.
Cut down on material waste. AI can analyze data on energy consumption, raw material waste, and other parameters and put forward recommendations on how to improve the manufacturing process. Also, the technology can predict demand, so that pharma manufacturers avoid producing large quantities of drugs that will not be consumed and will otherwise go to waste.
The pharma sector largely depends on sales. Companies aim to reach as many customers as possible while offering a distinctive user experience and a customized approach. Artificial intelligence in pharma can facilitate drug marketing by:
Comparing past marketing campaigns and identifying the most profitable approaches. The technology can also analyze customer engagement tactics and pick the most successful ones.
Aggregating customer data in real time to understand their behavior and what they are looking for to create a tailored advertisement.
Optimizing pricing of new drugs by considering all involved stakeholders and data on similar medications.
Simulating different market scenarios by forecasting changes in demand, competitor behavior, etc. This allows pharmaceutical firms to be prepared for sudden landscape changes.
Finding new consumers for existing drugs. For instance, Pfizer relied on AI to spot and reach new potential customers for Chantix (a drug that helps people quit smoking). The tool analyzed data from the Centers for Disease Control and Prevention to identify previously untapped population segments.
Drug dosage optimization
AI can analyze large quantities of unstructured patient data and calculate the optimal dosage of a particular drug for this person to achieve the best possible results with minimal side effects. Artificial intelligence models in the pharma industry can analyze the following information:
Medical history, such as doctor’s notes, lab test results, genetic makeup
Medical images, such as Magnetic Resonance Imaging (MRI) scans
Biomarkers, such as protein levels and genetic mutations
Drug characteristics, such as its metabolism
Potential side effects of a drug and of similar drugs
When the optimal dosage is calculated, the technology can monitor its effectiveness and make adjustments when needed.
To give a real-life example, a California-based company Dosis built an AI-driven personalized medicine dosing platform that dialysis clinics can use to manage chronic drug intake. In his interview with HealthcareITNews, Dosis’ CEO Shivrat Chhabra mentioned this platform helped clients reduce drug consumption by 25% while improving patient outcomes.
Challenges associated with implementing AI in pharma
Some of these obstacles are specific to the field, and some are more general and apply to all projects involving this technology. One of the key challenges is the enormous costs associated with artificial intelligence. This is particularly hard as the expenses associated with drug development are already rather high. You can turn to experienced AI consultants to learn how to cut down on costs and still get a viable product.
Here are other prominent challenges that you can face during pharmaceutical AI implementation.
Data quality and quantity
According to a recent study by McKinsey, the lack of integrated data sources was the chief obstacle on the way to applying analytics in the healthcare field.
Pharma AI models typically require large datasets to learn. However, it’s a challenge to obtain a sufficient dataset for each disease, especially the rare ones. So, as training datasets are getting smaller, the data that an AI-powered drug development tool has to handle is rather complex. Think of patient data. It includes historical information, genetic makeup, doctor notes, medical scans, etc. Under these conditions, it’s a challenge to build accurate algorithms.
When training data is lacking, it’s possible to use synthetic data generators for some pharma applications. For instance, Mostly AI claims it can generate data suitable for pharmaceutical usage. Healthcare data is among the most sensitive data types, and privacy is of the essence in such applications. Synthetic datasets can solve this issue. As Andreas Ponikiewicz, VP of Global Sales at Mostly AI, puts it, “With generative AI based synthetic healthcare data, that contains all the statistical patterns, but is completely artificial, the data can be made available without privacy risk.”
Another option for acquiring data for experimenting with AI and pharma is to become a part of a specialized collaboration. For example, the Massachusetts Institute of Technology initiated the Machine Learning for Pharmaceutical Discovery and Synthesis Consortium. 13 pharma companies joined the consortium to design and build AI algorithms for small molecule discovery.
You need to make sure that the data used within pharmaceutical applications is all realistic. But it’s rather costly to verify that, as it requires the intervention of human experts.
Lack of interoperability and a unified data standard
There are still multiple healthcare IT standards and regulations, which means that each hospital can adopt a standard of their choice for data storage and formatting. This makes it hard to integrate and use patient data needed for drug-related research from different medical facilities.
These issues of AI in the pharma industry can be addressed on the governmental level. For instance, the Swiss Personalized Health Network (SPHN) is a health data unifying initiative by the Swiss government. The SPHN was set to build a national infrastructure that streamlines medical data exchange among Swiss hospitals, research institutes, and regulatory bodies.
On an individual level, pharma researchers can benefit from platforms like Deep 6 AI, which uses NLP to scan and extract data from heterogeneous electronic health record (EHRs) systems.
“All data is biased. This is not paranoia. This is fact.”
– Dr. Sanjiv Narayan, professor of medicine at Stanford University.
AI-powered models can easily develop bias if their training dataset wasn’t representative of the target population. Data bias has specifically been a problem in the pharmaceutical and healthcare sectors. Research shows that only a few AI-powered products submitted for FDA approval offer evidence on covering the bias issue.
Some medical professionals believe that it will help reduce bias if data scientists work more closely with clinicians and learn more about data while building the algorithms. They can request information, such as where the data came from and what was the original goal of gathering it. Then engineers can make tweaks to the algorithms to address any population misrepresentation.
Algorithms can also acquire bias as they continue to learn on the job. Hence, systematic audits are essential to ensure that all AI-based tools are still relevant and work as expected.
Integration with existing systems
Deploying AI in pharma implies integrating it with the existing platforms and applications. Many pharma companies still rely on outdated legacy systems that are not designed to work with AI or deal with a large amount of data. Such systems use their proprietary protocols and are hard to integrate with modern applications.
Pharma companies that want to use modern technology alongside legacy systems can benefit from custom pharma software solutions designed to fit seamlessly with the existing legacy systems.
The complexity of the pharmaceutical applications
The use cases of artificial intelligence in the pharmaceutical industry are rather complex, and there is a large room for error in the predictions that the technology makes. Here is what makes pharma so intricate:
Every patient has individual characteristics and many factors to consider in clinical trials. If you are developing a drug for liver-related issues, you need to find trial participants with no other health conditions that can influence and sway your results.
The need to consider the interaction between different drugs as one person might be taking multiple drugs to treat different conditions.
Disease variability as one medical condition can have several variants and manifest itself in different ways.
Training datasets describing diseases and treatments are not balanced, which can force the algorithm to recommend the most frequently occurring solution even if it’s not the correct one.
To sum it up
Deloitte reports that only a few of the 7,000 rare diseases that we know have witnessed some progress over the past years. And the consultancy believes AI in pharma can change this. In addition to the applications mentioned above, AI can help pharma companies achieve compliance, which is vital in this field.
If you want to incorporate this advanced technology into your business, you are likely to have to team up with a tech vendor of your choice. Also, it’s a good practice to:
Make sure your training dataset is realistic, even if the verification process is expensive and requires human expert intervention
Incorporate AI into your strategy instead of treating it as a side project
Build strong AI skills or outsource this to dedicated teams
Encourage a close collaboration between your data scientists and clinicians
Beware of the latest regulations regarding using AI in pharma, as these are changing rapidly
Build your own ethical expertise to address any concerns associated with AI and pay a special attention to privacy and security if you decide to collaborate with other players in the field and share your data
Regularly monitor the algorithms’ performance for bias and inaccuracy, be it for disease discovery, recruitment of trial participants, or even drug advertising. For example, the University of California conducted a study on advertising mental health medication on social media and discovered that AI models tend to excessively recommend these drugs to Latino and African-Americans customers.
When using AI to generate chemical compounds, always validate the results, as it can deliver toxic or otherwise unsuitable components
Speaking of the future of artificial intelligence in the pharmaceutical industry, PwC predicts the emergence of a new digital health ecosystem that will include the following players:
Solution vendors, who will offer personalized treatments, drug dosages, etc.
Orchestrators, who can use AI and analytics to address patients’ needs
Platform providers, who will mediate between the aforementioned players
And according to the consultancy, firms who will still refuse to make AI a part of their operations will turn into a mere “contract manufacturers” for the rest of the ecosystem. So, if you haven’t yet considered enhancing your business processes with AI, this seems like a good time to experiment with the technology.