Q: How have evolving technologies like artificial intelligence (AI) and machine learning (ML) changed how you conduct research, manufacturing, or business?

Q: How have evolving technologies like artificial intelligence (AI) and machine learning (ML) changed how you conduct research, manufacturing, or business?

June 15, 2021PAO-06-21-RT-1

Evolving Technologies

Stephan Rosenberger, Ph.D., Head of Digital Transformation, Lonza

A: Artificial intelligence (AI) nowadays represents the cross-industrial dream of acceleration through automation and precision. This near-magic set of technologies has already proven its value in highly regulated industries and, in particular, is establishing itself in the pharmaceutical landscape. The increasing manufacturing complexity of new medicines and the desire to reduce “time to market” are driving faster development and implementation of AI solutions in our activities.

Indeed, we see improvement through AI resulting in shorter cycle times in our operations while increasing quality and/or reducing overall costs and raw material consumption — an additional step toward sustainability. 

In particular, we successfully optimized product quality using computer vision technologies in quality assurance and are taking our first steps towards AI-enhanced genetic engineering based on bioinformatics methods. One of our latest endeavors is to develop hybrid approaches leveraging AI, mechanistic models, and traditional statistics for process scale-ups.

Nevertheless, AI is not the only answer to new challenges and is often only one part of the puzzle, though a pivotal one. In practice, already established methods still deserve their place in innovation next to AI, supporting human intelligence. Reaching the ultimate goal of lights-out manufacturing — closer and closer to the vision of artificial general intelligence — will still require tremendous effort and a substantial boost in IT infrastructure.

At present, AI already provides a strong basis for greater automation and knowledge expansion as an enabler of better, faster decision-making. We believe the industry is at a critical stage in our digital transformation journey.

David Vodak, Ph.D., Senior Director, Head of Small Molecules R&D, Lonza

A: With respect to research, which impacts manufacturing and business operations generally, we have pursued machine learning (ML) to leverage data for optimizing process engineering and formulations to arrive at more robust solutions more efficiently for our clients.  

The pharmaceutical industry is data-rich, yet these data are generally not used as effectively as they could be, so we are working to make use of data to be more predictive of technical outcomes to give us better results more quickly. This is important, as a large fraction of new chemical entities in the industry are being advanced under accelerated development programs, such as breakthrough therapy designations, where speed is critical to bring new medicines to patients. 

In addition, the complex nature of the chemical matter (modality, size, solubilities), as well the need to deliver these new modalities effectively into the body with speed and agility in the development program, adds an extreme need to be predictive and use all of the data that came before to be successful.

Artificial intelligence (AI) and ML are likely going to be cornerstone technologies in drug development of the future.

Wolfram Schulze, Vice President Information Systems & Organization, Rentschler Biopharma SE

A: The application of evolving technologies like artificial intelligence (AI), machine learning (ML), and mixed reality (MR) is an important pillar of our digital agenda at Rentschler Biopharma. Hence, these are an integral part of our planning for strategic expansions and new services in the future. It is no secret that innovative technologies provide us and our clients a competitive edge in our sector by enhancing traditional services by digital means. I believe that — against the background of continued democratization of healthcare — this new dimension of digital maturity is becoming increasingly important in the CDMO selection process, beyond pure manufacturing capabilities. As a premium CDMO, our strategic vision drives us to not only offer exclusive services but also make digital differentiating capabilities available to our clients.

Alan Stevens, Ph.D., Senior Principal Scientist, CatSci Ltd.

A: At CatSci, we accelerate the development of the processes used to make the medicines of tomorrow.  We are increasingly seeing opportunities to use high-quality data sets to predict outcomes, supplementing our use of theory-driven models. There are now browser-accessible tools, trained using test data calculated computationally, that allow our chemists to establish bond-dissociation energies, redox potentials, and other molecular properties. 

Our collaborators in the areas of catalysis have used supervised machine learning to predict outcomes in “workhorse” chemical transformations like C–N couplings, Suzuki reactions, and C–H activation transformations. This has involved feeding a neural network model with calculated descriptor data for reactants and reagents, together with test experimental data.   

As well as being informed around the likely outcome when using certain reactants and conditions, the broader suite of computer-aided synthesis planning tools is increasingly catching up with chemists in its ability to establish the most suitable way of putting together a target active pharmaceutical ingredient.  These tools also inform around the choice of reagents and solvents used in process design work packages. If nothing else, they bring a different set of perspectives to the table compared with an experienced scientist! 

From an analytical perspective, algorithms trained using years’ worth of chromatographic retention time data, together with 3D structural information for the analytes, are enabling the prediction of retention times for new compounds.  While our leveraging of machine-learning methods is likely to increase going forward, we cannot subcontract too much to a data-driven algorithm. Knowing how it works is still required so we know when we are being misled!

Valerie Van Hulle, Global Strategic Marketing Manager, BASF Pharma Solutions

A: Evolving technologies have transformed BASF’s ability to offer new services and solutions to our customers in the pharmaceutical industry. From a service perspective, one example is BASF’s ZoomLab™, an online virtual formulation assistant that allows formulators to predict their next starting formulation for a pharmaceutical product and solve other formulation challenges. In terms of technology, ZoomLab™’s soon-to-be-released chatbot uses artificial intelligence to assess a formulator’s intention and provide personalized recommendations to specific formulation challenges. Additionally, several ZoomLab™ features, including the Decision Support for Solubility module, were developed via machine learning. Because of developing technologies, BASF will be able to continue to advance digital services such as ZoomLab™.

Digitalization has also enhanced the solutions we can provide to our customers. BASF’s R&D teams are exploring machine learning in molecular modeling to predict trends in experimental data. Our manufacturing and supply chain teams are leveraging “big data” to streamline the production of our active ingredients and excipients. We know there are many opportunities to leverage digitalization, and we will continue to invest in advancing technologies.

Behzad Mahdavi, Ph.D., MBA, Vice President of Open Innovation, Biologics, Cell and Gene Therapy, Catalent

A: Technologies such as AI and ML are poised to transform current pharmaceutical approaches, with the potential of increasing productivity, ensuring robustness, reducing costs, and accelerating timelines, all of which are key objectives of the industry. The advantages of AI are the detection and extraction of information from patterns in complex data sets that are difficult to observe conventionally and to automate actions based on that information. ML provides the ability to “learn” by analyzing data over time.

The application of AI and ML technologies in the biopharmaceutical industry is still in its infancy, and they are mainly used where GMP requirements do not apply. A progressive and phased approach in their implementation is crucial, so that knowledge and experience gained in the early phases can be incorporated into systems for the later phases. Biopharmaceutical process development can be highly complex, with multiple interactions potentially affecting the critical quality attributes of products. As biomanufacturing evolves from batch processing to a more continuous operation supported by inline sensors and process analytical technology (PAT), AI/ML could be naturally introduced alongside, leading to data-driven continuous operations that could significantly improve product quality, reduce production costs, and shorten the time to market.

To ensure support from regulators, these processes need time to demonstrate their robustness and compliance within the GMP guidance. Initial examples, such as AI-based real-time maintenance or the use of imaging in quality control or cell characterization, could potentially help to build some level of basic confidence.

Matthew Bio, Ph.D., President & CEO, Snapdragon Chemistry, Inc.

A: Even simpler than technologies like AI and ML, access to low-cost computing power, IoT and automation technologies have dramatically changed the way we do research and process development. Early in Snapdragon’s history, we invested in developing a lab operating system that would allow us to communicate with all manner of equipment. LabOS is particularly useful for continuous manufacturing process development. IoT integrated into LabOS allows supervisory control and data acquisition of all electronic lab equipment via a web-based interface. LabOS is continuously gathering data from in-line sensors and PAT for continuous process monitoring.  With all components of the system connected, advanced automation is enabled, allowing for scripted automated experimentation.  With programmed safety interlocks and automated text message alerts, unattended and remote operation of lab equipment can be done routinely and safely.  Moving into simple AI applications, Snapdragon has created an algorithm-driven autonomous optimization platform connecting automated reactor systems to online analytical systems (e.g., UHPLC).  The algorithm will search multivariable landscape optimizing for an objective function, such as yield, which can be read out from the connected PAT. We are hopeful that the data gathered from these systems can subsequently be used in ML applications to accelerate reactions and process development.

James Rogers, Head of Manufacturing and Supply, Sterling Pharma Solutions

A: For contract development and manufacturing organizations (CDMOs), investment in technology is crucial to ensure that capabilities are in place to respond to customer demand and the forecasted nature of future projects. Whereas previously, the primary consideration for innovators seeking partners was the lowest cost, for many companies accelerated program development is now crucial, so minimized lead times are preferable. This means that CDMOs are looking to invest in technologies that can reduce the project timeline while at the same time enhancing manufacturing efficiency and minimizing costs, as well as ensuring product quality.

Digitization and machine learning strategies are technologies that are becoming more important, and they can play a key role in reducing the timescales of projects. At Sterling, we are working with partners to develop a solution that will allow us to monitor process parameters during manufacturing and identify as early as possible if there is any deviation from pre-set limits. This should ultimately allow appropriate steps to be taken to prevent any event that might lead to a product becoming out of specification, and reduce the risk of unexpected increases in costs.

This approach differs from traditional process control networks and could enable the integration of historic data and events into systems — so that any actions taken will factor these in — and thus allow faster and more accurate interventions. Such technology is core to supporting operational excellence by reducing variation in and between batches of API and optimizing the quality of customers’ products.

William Greene, M.D., Chief Executive Officer, Fountain Therapeutics

A: Developments in artificial intelligence (AI) and machine learning (ML) over the last decade are, in fact, one of the two pillars that made the origin of Fountain possible. The other is our experienced team of biologists and computer scientists, who turn large data sets into biologically meaningful information for the development of therapeutics for age-related diseases. Research confirms that aging is the major risk factor of disease and that the aging process is, in fact, malleable — it can be delayed and reversed. Diseases of aging can involve multiple biological pathways shared among different cells, tissues, and organs. Yet, traditional target-based drug discovery approaches do not adequately identify the pathways and molecules responsible for cellular aging and its reversal, since they assume single pathway malfunction and look for known biological pathways of disease. Advances in AI and ML have enabled Fountain to spearhead a new unbiased approach to drug discovery and development. We have built an AI-based platform and ML-driven model using computer vision that can predict the biological age of cells with unprecedented precision by discerning patterns in young versus old cells based on visual features. This highly scalable platform enables us to screen thousands of molecules and discover those with the ability to produce younger, rejuvenated cells, without focusing on specific pharmaceuticals targets, at least at first. This unique AI-driven platform, in combination with our highly parallel in vivo disease testing, positions us to develop a unique pipeline of truly disease-modifying therapies for age-related diseases.

Nicolas Poirier, Ph.D., Chief Scientific Officer, OSE Immunotherapeutics

A: Drug development, especially for immunotherapies used in cancer and for autoimmune diseases, is an extremely challenging process with an incredible number of variables at play. In addition, modern immunotherapies are often large molecules, such as biologic drugs modified from proteins found in nature, which adds additional obstacles for early clinical testing and high-throughput screening of drug candidates. Together, this makes early-stage drug development — especially preclinical screening of potentially valuable new therapies — a costly endeavor fraught with failure in more cases than we see success.

As a result, in the past few years at OSE, we have incorporated artificial intelligence (AI) and machine learning algorithms into our early screening of new drug candidates. In early 2020, we initiated a partnership, which was expanded in early 2021, with a leading developer of AI-based software for antibody discovery and development, MAbSilico, to help accelerate drug development efforts, including improvements in the selection of a promising drug target to the optimization of the final humanized lead.

MAbSilico's software platform, in combination with OSE's expertise in immune-relevant antibodies, allows for rapid in silico screening of potential candidates. A great example of this in action is OSE's CoVepiT, a prophylactic multi-target vaccine against COVID-19. AI-assisted peptide modeling identified vaccine target epitopes in only two weeks, helping OSE move quickly into preclinical and human ex vivo studies in August 2020. We are excited to continue working with the MAbSilico team and their AI-software platform to help identify the next round of "Immunotherapy 2.0" drug candidates.

Michael Vidne, Ph.D., Chief Business Officer/Chief Strategy Officer, Fore Biotherapeutics

A: At Fore Biotherapeutics, we are leveraging our functional genomics platform, Foresight, to match populations of patients with rare cancer mutations to targeted therapies that have a high probability of making a difference. Our mission is to provide patients with unaddressed cancer mutations with rapid access to the right medicines.

Advances our team has made in machine learning, specifically in deep neural networks, allow us to analyze images of billions of cells emulating the patients’ mutations and assess their response to different targeted therapies.

By conducting functional genomics screening of rare cancer mutations and analyzing the data, the Foresight engine can identify targeted therapies that could serve patients who might otherwise never have found a clinical match. We then in-license those therapies to guide them through clinical development, with the ultimate goal that hundreds of thousands of patients could receive precisely tailored, previously untapped treatments.

It’s a new kind of precision oncology model that would not have been possible without the tools that are now available in AI and ML. We hope that, as we continue to advance the Foresight engine, we can guide additional technological progress that will enable faster development of and access to precision cancer treatments for years to come. 

Eric Richman, Chief Executive Officer, Gain Therapeutics 

A: Evolving technologies are the foundation for our work in oncology, rare diseases, and neurodegenerative disorders. The power of supercomputing has paved the way for new methods of drug discovery. We are redefining drug discovery with our SEE-TxTM platform technology, designed to discover new allosteric binding sites on misfolded proteins and predict their druggability. Protein misfolding is the root cause of many diseases with high unmet needs such as Parkinson’s disease, cystic fibrosis, Gaucher disease and other lysosomal storage diseases.  Advances in computing have enabled us to redefine drug discovery by targeting the cause of the disease rather than simply treating symptoms.   

SEE-TxTM utilizes the known 3D structure of proteins and supercomputing technology to find druggable binding hotspots on misfolded proteins and correct their shape. This has the potential to up- or downregulate enzymatic activity, causing the protein to gain or lose function and eliminating the subsequent toxic substrate buildup that causes disease. The platform utilizes a unique algorithm, based on a patented method to analyze molecular dynamics. Once a druggable binding site is found, a separate algorithm is then used to scan libraries of small molecule compounds to identify those that have the potential to bind to the site and restore function. This offers a new drug discovery approach that is differentiated, highly specific, efficient, and cost-effective. Through SEE-TxTM, we look forward to changing the way the industry thinks about drug discovery and, thanks to evolving technologies, are one step closer to getting novel treatments to patients in need.

Fiona Law, Partner and Patent Attorney, Potter Clarkson

A: For us in the IP profession, we’re beginning to see some AI integration into our work, but we’re still a long way off from it becoming a major part of our daily lives. AI is being used by some patent offices for more efficient searching, for example, and I can see how AI might assist in drafting legal documents. When devising an IP strategy for a startup or conducting complex IP due diligence, there’s simply not enough data to train AI algorithms on, and it cannot replicate years of experience.

The story is very different in the healthcare industry, as many of our clients in the sector are feeling a profound effect from the use of AI. Crucially, unlike in other industries, data is readily available in the healthcare sector. Patient data, especially in countries like the UK that have large healthcare systems, is easier to obtain and broad enough to train algorithms on for new product and treatment development. AI is becoming increasingly important in the personalized medicine and diagnostics space.

AI is also revolutionizing drug discovery. Traditionally, hundreds of researchers would spend hours developing and testing new drugs, and there was a high failure rate. By using AI to quickly sift through data to find possible treatments, healthtech startups are able to massively decrease the time taken and cost of developing new treatments. In vitro and in vivo data is still required of course, and the role of the pharmacologist is still very important, but it’s undeniable that AI is having an impact. Big pharma is recognizing this, and we’re seeing more partnerships forged between them and healthtech startups.

Read Part 2: Social Media Marketing
chat button