Technologies Transforming the Pharma Value Chain: Part 1

Like most other industries on the planet, the pharmaceutical industry is in the midst of a profound transformation as artificial intelligence (AI) is brought into the fold. AI, which refers to the science and application of algorithms and computer programs capable of exhibiting characteristics of intelligence, like learning, problem-solving and decision-making, is producing changes up and down the entire pharma value chain, from drug discovery and clinical trials to marketing and strategic planning.

This article provides an overview of AI solutions being used by pharmaceutical companies today to develop new and better molecules and bring them to market more quickly and effectively. Part 1 will explore technologies used from drug discovery to clinical trials, regulatory and supply chain, and Part 2 will explore technologies in market access, strategic planning, market research, and sales and marketing.

Technologies used in drug discovery

Covid-19 offered a perfect storm in which to test AI to speed up the identification and development of drugs, both for treatments and vaccines. Several AI-powered companies led this pack, revolutionizing the drug discovery phase, which, with traditional methods, can take as long as 10 years and cost upwards of $2 billion.

For example, South Korea-based Deargen used AI to examine the likelihood of existing molecules to bind and block a prominent protein on the outside of SARS-CoV-2 and identified two existing drugs to help combat Covid-19, as well as several new compounds. Benevolent AI used a similar approach and found 6 molecules, including Ruxolitinib, now in clinical trials.

Rather than seeking to repurpose existing drugs, Insilico Medicine used AI to generate tens of thousands of novel molecules with the potential to bind to specific SARS-CoV-2 proteins, a list which was then narrowed down to just 100 molecules after 4 days with the help of deep learning (DL) algorithms. SRI Iktos also made use of DL to virtually craft new molecules, while SRI Biosciences AI-powered virtual chemistry platform is used to figure out the best way to create novel compounds. Together, they can design, make and test new molecules in as little as one to two weeks.

Finally, Moderna has integrated AI into every step of its drug discovery and manufacturing process. In mere days after Chinese authorities released the Covid-19 sequence, they finalised the sequence for the mRNA vaccine, and were able to release a first batch in early February, 2020, with clinical trials starting in March, 2020—for a process that historically took years.

These few examples provide ample evidence of the benefits of using AI in drug discovery. As for specific processes, there are a number of ways AI can be applied to this domain.

AI is able to rapidly and intelligently aggregate, organize and analyze huge amounts of data from and in a wide variety of sources and formats, leading to quicker discoveries of new drugs and a clearer understanding of the mechanisms of a disease and its influence on phenotype and pathology. Neural networks (NN) and DL, a form of representational learning where iterative and increasingly complex models allow for high levels of abstraction without the need to train the algorithm, are particularly useful in this regard.

It enables companies to not only find new uses for existing drugs, but also to generate new drug candidates (by, for example, analyzing RNA data from patients to identify new biomarkers and drug targets), validate drug candidates (by understanding how the drug will interact with the body for greater safety, stability and efficacy), and drug design (for example, making and modifying DNA to enable prototyping and editing of recombinant molecules for vaccines).

AI and DL are also being used to make strides in precision medicine, which examines the pharmacogenetics, environment and lifestyle of an individual to prevent, diagnose and treat disease. Deep learning algorithms “have been shown to make diagnoses at least as well as physicians in cardiology, dermatology, and oncology,” and can provide success rates upwards of 99.5% when used to support human pathologists’ diagnoses.

A variety of pharmaceutical companies are using DL and NN to further precision medicine today, such as Deep Genomics, whose “new generation of computation technologies … can tell physicians what will happen within a cell when DNA is altered by a genetic variation, whether natural or therapeutic”, and Atomwise, whose AI technology enabled them to find two existing drugs which may be effective in combating Ebola.

Drug discovery is a critical step in the pharmaceutical pipeline, but has historically been extremely costly and time-consuming, with only roughly 30% of pre-clinical studies entering Phase 1 of clinical trials, and only 1 in 10 of these will ever reach the market. As AI continues to develop in sophistication and use-cases, we can hope to see far better results in the near future.
Technologies used in clinical trials
According to ScienceDirect, only 14% of clinical trials meet patient enrolment deadlines. 80% fail to finish on time, and at least one fifth are delayed for six months or more. 85% fail to retain enough patients. And yet, the cost of clinical trials is roughly $7 billion per year, representing nearly 40% of the United States’ pharmaceuticals research budget.

AI is being used in a variety of fashions to reduce costs and preclinical and clinical delays and increase success rates. AI can be used in the design of preclinical experiments by, for example, analyzing data on reagents like antibodies and presenting published figures with actionable insights to reduce time, money and uncertainty in planning experiments.

Likewise, the actual operation of preclinical experiments can be improved with AI. Automated labs, for example, represent an inexpensive, on-demand resource for pharmaceutical companies that can greatly reduce the costs in time and money of preclinical laboratory work.

AI has much to offer where recruiting is concerned, too. Advancements in natural language processing (NLP) has allowed for highly sophisticated “chatbots” that are capable of recruiting and enrolling eligible patients with little oversight from human teams. AI can also be used to predict outcomes based on patients’ electronic health records (EHRs), for more rapid and accurate screening.

In one Phase 2 clinical trial for prostate cancer, for example, Eularis was able to use AI to identify and recruit 1,739 eligible patients when approached by a team whose initial efforts had resulted in just six—and this in just 10 minutes, from EHR, claims and AMA data. In another case, a pharma company running a third-line cancer product trial had many patients dying at 2L (such that 3L patients were rare and difficult to find). Leveraging big data and AI, Eularis was able to identify and examine all previous 3L patients to predict which patients were most likely to progress to 3L, creating a visual presentation of potential patient locations with physician IDs and other data, enabling the sales rep to prioritize who they saw.

Advancements in wearables, many of which now feature quite robust on-device AI and machine learning (ML), also make it easier to monitor trial patients and obtain far richer and more granular data, including heart rate, glucose levels, movements patterns, and more. Patients can be monitored remotely, and a single team, leaning on AI to organize and analyze all the data, can effectively oversee a large number of clinical trials. In a similar vein, the advent of virtual clinical trials, in conjunction with wearables and smart devices, likewise allows for a larger number of concurrent trials than ever before.

Of course, running clinical trials generates a significant amount of documentation and reporting, which, when done by humans, can be costly and particularly time-consuming. For example, one Eularis client needed to prepare 2000+ reports for disclosure by removing all personally identifiable information (PII), personal data (PD) and company confidential information (CCI). Offshore outsourcing was slow and of poor quality, and made for difficult accuracy assessment. An AI approach was built to automatically identify PII, PD and CCI, and a full redactio was completed of the 2000 documents in a two-year time period, running at a pace of between 20 to 40 units per week.

Finally, writing up clinical study reports (CSR) can take internal teams up to 2 months, with every day getting the drug to market costing companies money. One solution is to create an AI to generate a regulatory-compliant report that is 80-90% complete in just hours. Eularis offers one such solution, and has been used to reduce internal resource use cost by approximately 250 hours and roughly $25,000 in team time, and saved $9 million per day by 6 weeks, adding $378 million to the bottom line.

Technologies used in regulatory

The pharmaceutical space is a tightly regulated one, placing significant stress and risk on pharmaceutical companies. AI excels at the kind of laborious, manual work hitherto performed—more slowly and less accurately—by humans, like collating daily guidance and regulations and monitoring news networks and journals. Likewise, much of the data processing work can be done by AI.

As already mentioned above, redaction of personally identifiable and corporate data from documents can be done at a much greater pace by an AI, and fully compliant CSR can be generated, 90% complete and in just one hour.

Regulations are complex and ever-shifting, but ultimately amount to nothing more than a series of intersecting rules, which AI algorithms have no problem parsing and applying to documents and data. It can also be used to scan (and render searchable) audio, image and video files, including text in images and videos or dialogue in audio that may pose a regulatory problem. This, in addition to always-on news monitoring, greatly reduces risk.

Technologies used in Supply Chain and Distribution

Pharmaceutical supply chain and distribution is among the most complex on the planet. In addition, a McKinsey study provided evidence of a profound lack of agility in pharma supply chains. Here, AI is already having an enormous impact, helping reduce costs and improve movement of molecules up to, during, and following production in the very few taking advantage of it in supply chain.

One way AI can help pharma companies improve their supply chain is by consolidating data from a variety of sources, organizing and cross-referencing it to provide firms with a 360º view of the conditions surrounding deliveries and movement. This includes data related to the manufacturing, internal and external sources, social media, trends in patient health, and even weather patterns.

Merck KGaA, for example, has used harmonized, real-time information, with everything from supply chain performance to stock-keeping units to data collected from the company’s ERP, to optimize operations. Sensors placed throughout its supply chain gather data on inventory distribution and availability for every SKU, providing end-to-end visibility that helps them process orders in a much shorter time. Merck’s AI is also able to forecast more accurately than humans.

AI can furthermore be used to automatically generate a “supply chain map”, including order details like allocated quantity and expected delivery date. Such a system can then deliver highly accurate recommendations and predictions based on ML, as opposed to simple rules-based available-to-promise (ATP) calculations.

As highlighted in an article from Forbes by Gary Hutchinson, for example, DataRobot is one solution that enables supply chain managers to “build a model that accurately predicts whether a given drug order could be consolidated with another upcoming order to the same location or department.” Analytica decision making takes this one step further, allowing real-world, real-time data, risk assessments and re-planning conditions to factor into such recommendations.

AI can also greatly streamline inventory management by determining which products are most likely to be needed (and how often), track delivery to patients, and monitor for delays or incidents so that replacements can be shipped out within hours, if needed. Warehouse automation also “speeds communications and reduces errors in ‘pick and pack’ settings. At its simplest, AI predicts which items will be stored the longest and positions them accordingly.”