In the pharmaceutical industry, R&D budgets typically track proportionally to revenue, so when revenues start to shrink, innovation is impacted. But when innovation budgets shrink, the challenges driven by pharma’s complex and lengthy product development processes—such as drug discovery, screening, and clinical research—remain.
Companies can respond to these economic challenges by exploring and adopting digital technologies to help improve top- and bottom-line growth by reducing the need for clinical research, speeding promising therapies to clinical trials, improving operational excellence, and reducing risk.
In this article, we will highlight some of the emerging value cases for artificial intelligence (AI) across phases of the pharmaceutical value chain, which includes processes in four primary functions across the pharmaceuticals lifecycle—Discover, Create, Make and Sell.
There are many challenges inherent to the drug discovery process such as understanding the relationships among the sequence, structure, and function of biopolymers. It’s also difficult to predict their interactions—both with one another and with small organic molecules that are native to the body or designed as drugs.
These problems are at the heart of genomic analysis, drug design, and protein folding predictions, and create computational complexity that is far beyond the human capabilities of classical problem solving. This is where AI-enabled Computer-Aided Drug Discovery (CADD) tools have huge potential to complement human decision making, reducing the time and cost associated with discovering new drugs.
Value Cases for AI in the Discover Phase
Identify therapeutic drug candidates by analyzing potential hit/lead compounds for optimized pathways. By predicting the mode of compound interaction with multiple targets, companies can develop a product that balances efficacy, quality and safety.
Develop novel biological products based on protein folding predictions. Proteins are made up of hundreds or thousands of amino acids, and these amino acid sequences specify the protein’s structure and function. AI can unlock this complexity by analyzing the enormous amounts of data collected within the genome database to learn the complex design rules of a protein’s structure and function. This learning can then be applied to generate an artificial protein structure which can rival those found in nature.
Identify biomarkers to enable precision medicine therapies by analyzing genomes and linking them to outcomes.
Apply text-based analytics leveraging natural language processing (NLP) to sift through the research literature, laboratory notebooks, and clinical trial datasets to find relevant insights from unstructured data sources, speeding up the discovery process.
Managing the quantity and complexity of data generated during development of new products has always been a challenge for pharmaceutical manufacturers. Most of this product development data is managed across enterprise systems such as a product lifecycle management (PLM) system, quality management system (QMS), laboratory information management system (LIMS) and clinical trial management system (CTMS).
Value Cases for AI in the Create Phase
Apply product lifecycle intelligence (PLI) to predict outcomes from process and recipe changes in PLM by analyzing historical samples from LIMS, quality incidents from QMS, related data for safety, efficacy, and toxicology, and related electronic batch records (EBR). This helps predict and mitigate likely change management risks before a design is transferred to manufacturing.
Use AI and PLI to identify data failure patterns in design data across the enterprise data landscape, effectively reducing the frequency and cost related to errors throughout the design process.
Leverage NLP and robotic process automation (RPA) tools extract reusable digital data artifacts from structured and unstructured data to for process and material definitions, shortening the latencies of Tech Transfer and Scale-Up and enabling a continuous digital data flow from IT systems to OT systems.
Pharmaceutical manufacturing performance depends on the quality and relevance of measurements for process monitoring and control. As biologics become more sophisticated, production processes become more complex, requiring the integration of a growing number of devices and control process parameters.
Enabling smart connected assets via IoT is critical to continue to effectively monitor the devices. When manufacturing processes become too complex for standard PID control models, AI can be applied as a supplement to existing control schemes, allowing manufacturers to intelligently optimize their processes.
Value Cases for AI in the Make Phase
Develop soft sensors to improve the control of non-linear processes. Many process parameters (such as temperature, and pressure) cannot be measured accurately because of non-linear process variations observed within complex physical processes and equipment settings. Soft sensors have emerged as an effective alternative to traditional hardware sensors for collecting, monitoring, and controlling critical process variables. This AI-enabled, data-driven approach offers a new and effective way of obtaining accurate readings along the entire processing chain without installing expensive hardware and can effectively improve efficiency and reduce scrap through enabling process control optimization.
Apply predictive maintenance analytics to manufacturing equipment to predict what will go wrong and when. Provide real-time decision support by looking across large data sets to find correlations between obscure signals and events and alerting users to potential failures before they happen.
Optimize the biologics supply chain itself—from formulation to manufacturing, shipment, and ultimately transportation to pharmacies, hospitals, and even homes. Machine Vision capabilities can train a model to assess shipping labels for errors and track serialization codes to prevent counterfeiting. AI can predict when a product is most likely to be needed, track exactly when it's delivered to a patient, provide delivery time estimates, and even track delays or incidents that may trigger a replacement shipment. This effectively optimizes order replenishment, preventing drug shortages and stock outs.
Reinforce the pharmaceutical cold chain by using blockchain to store data gathered from sensors on IoT devices. This data can be used for more effective real-time decision making, and it can also be aggregated to generate predictive models based on environmental hazards throughout the cold chain cycle, helping to assess and address potential risks before the cold chain starts.
Soft sensors are computer algorithms (software and sensor combined) that perform calculations to generate extrapolated values based on the machine-learning model of a physical process. Soft sensors use process variables that are measured and recorded reliably online using available physical sensors or oﬄine through laboratory analysis results. Machine learning is applied to support soft sensor modelling in data-driven complex bioprocesses where it can represent nonlinear systems better by analyzing historical batch process data to generate predictive models while efficiently handling large datasets.
Pharmaceutical manufacturers are responsible for monitoring and reporting therapy effectiveness and adverse effects or side effects of pharmaceutical products on the market. Inconsistencies in adverse event reporting across the healthcare system present a huge challenge to pharmacovigilance. AI models have the potential to drive improvements by detecting adverse events directly from patient electronic health records and other primary sources.
Pharmacovigilance is the ability to assess and report the safety of the drug after it has been approved for use.
Value Cases for AI in the Sell Phase
Apply RPA to automate the manual and routine tasks associated with single case processing (collection, assessment, and reporting). This decreases costs and eliminates manual errors during pharmacovigilance.
Implement augmented adverse event intelligence by training machine learning models to extract and classify data directly from adverse event reports and predict the risk profile of incoming adverse events in real-time, notifying regulatory groups when needed. This automates drug safety analysis and decision support within pharmacovigilance, enabling a more responsive knowledge system for product benefit-risk management.
Identify clusters of selected patient profiles that are prone to developing an adverse event. In the long term, this will complement value-based reimbursement models with drug safety information.
Improve patient services programs by facilitating virtual nursing assistants, enabling connected applications, and analyzing the collected data.
Use NLP to analyze a broad set of text data from websites, social media feeds, direct consumer feedback, or electronic medical records to detect any unexpected benefits of a pharmaceutical product. This can lead to an expansion of indications for an already marketed product and provides an opportunity for pharmacovigilance to improve patient care.
Virtual nursing assistants are an advancement of telehealth services where a patient interacts with an advanced chatbot, answering questions about their symptoms. NLP mechanisms correlate symptoms and responds with specific advice and referrals to the appropriate healthcare provider
Getting Started: Key Considerations
Progress in computing hardware and algorithms will enable the transition many of these exciting value cases from potential to reality in the coming years. Leading life sciences and pharmaceutical companies have already started this journey by investing in AI and machine learning.
However, driving real value requires different ways of thinking, new and highly sought-after skills within the organization, distinct IT architectures, and novel corporate strategies. With the potential to lower cost, create new and effective treatments, and improve patient outcomes, AI is the future of pharma, but the technology is available now.
Those who get started now—when standards, strategies, use cases, and ecosystems are still being developed—will be ahead of the pack, but there are a few things to consider.
New technologies open up new possibilities, but enablement of the technology has to keep the end-user and experience in mind. For regulated industries, every piece of digital data related to development, manufacturing, and distribution is auditable by any number of regulatory agencies, so use cases must be well defined, verified, validated and completely transparent.
Three Steps for Getting Started
Identify and empower champions for AI and machine learning to explore potential applications and create pull across the organization.
Cultivate partnerships with the emerging artificial intelligence and machine learning ecosystem of like-minded research labs, academic institutions, technology providers, application developers, and start-ups.
Prioritize AI and machine learning use cases for small-scale proof of value (PoV) investments according to their potential for attaining business insights and value. When prioritizing, consider your organization’s therapeutic focus areas, business strategy, associated customer value propositions, and future growth plans. Monitor PoV performance and quickly scale those that prove most effective.
AI and machine learning should be core to your organization’s digital strategy. Moving forward, every business opportunity must be viewed through an analytics lens. This cultural and mindset shift will enable a data-driven organization that can nimbly identify, prove out, and capture value from opportunities.
Stuart enjoys travelling the world and tries to go where there is great scuba diving. Additionally, Stuart enjoys hiking in the mountains where he can unplug for a while and enjoy the beauty of nature. Stuart is married with two teenage daughters and resides in Austin, TX.