January 10, 2023 PAO-11-022-CL-15
The drug development process can be divided into three different stages: target discovery, candidate identification, and process development and scale-up for commercial manufacturing. Target discovery involves evaluating the genomics of a disease to determine how genetic mutations affect proteomics — the full complement of proteins produced in the relevant tissue. Enormous volumes of data are generated, much of which is not analyzed using conventional methods and are therefore not available for use, such as in correlating symptoms associated with the target disease to similar symptoms appearing in other diseases.
During the candidate identification stage, molecules that may potentially modulate the target identified in stage one are sought. That process involves extensive in vitro and in vivo screening of huge numbers of molecules, both of which are time-consuming and expensive. In addition, the identified molecules are often not easily translatable into clinical candidates.
On the manufacturing side, extensive process optimization is required to develop processes that are robust, efficient, and practically scalable. Formulation optimization is also necessary to generate a drug product with high efficacy that is safe to administer to patients.
At each of these three stages, human capabilities are limited to a certain degree. Artificial intelligence (AI) has the potential to expand those capabilities.
The main way in which AI is expanding those capabilities is by processing the unimaginably large data sets now being generated at all phases of drug discovery, development, and manufacturing. The quantity and complexity of data generated today are beyond the ability of humans to fully process, even with traditional computational tools.
Following the sequencing of the human genome, a transition began to advance the field from genomics to proteomics and metabolomics. These areas are all extremely data intensive. Conventional statistical methods are limited in their ability to extract valuable information. With AI models, machine learning, “high-powered gardens,” and other advanced digital tools, it is possible to identify relationships and uncover patterns and correlations that previously would have remained hidden.
Numerous benefits can be realized by applying AI and other digital tools — enhanced speed, reduced cost, lower workloads and thus labor requirements, increased throughput, and greater accuracy. Beyond these quantified advantages are less tangible but equally valuable benefits, including the ability to use types of data — for example, real-world data — that could not be adequately leveraged in the past.
Combined, these benefits of using AI contribute to a simplification of drug discovery, development, and manufacturing while also de-risking all three phases. Compound screening for candidate selection can be simplified using AI; in silico screening of large quantities of possible compounds using highly advanced predictive models (e.g., virtual cells, virtual animals) can rapidly identify those with the greatest likelihood of providing efficacy and safety and being industrializable. Only those compounds need to be subjected to in vitro and in vivo screening, dramatically reducing the time and cost of drug discovery while also increasing the chances of identifying candidates with high potential for success. We expect, in fact, that in the next five to eight years, approximately 20–30% of drug candidates reaching the clinic will have been designed using AI as part of the discovery process or in the preclinical development stage.
While the adoption of AI in the biopharmaceutical industry is a relatively recent phenomenon, there have already been many examples of its successful application. It is regularly used today in the recombinant protein and antibody space for sequence design, antibody affinity maturation, antibody humanization, and even manufacturing process optimization. AI also finds growing use in the diagnostics field, particularly for image analysis to determine whether cells are healthy or diseased and to detect various cancers. In these areas, AI can help companies overcome manpower and resource challenges and speed up the evaluation process.
Many success stories involve collaborations between pharmaceutical companies and AI software developers. The most well-known involve Big Pharma firms (e.g., Roche, Pfizer, GlaxoSmithKline) partnering with leading tech companies (e.g., IBM, Google). For example, Pfizer has partnered with IBM to advance its drug discovery process for cancer therapies.
The trend toward collaboration is a hallmark of AI adoption by the biopharmaceutical sector. Most major pharmaceutical companies are investigating ways to implement AI across drug discovery, development, and manufacturing, often in partnership with tech companies specializing in AI technology, from startups to well-established technology providers.
Some of the AI firms are focused specifically on pharmaceutical applications, while others are generalist AI companies working across multiple industries. Both approaches have their advantages for pharma and biotech developers looking for assistance with AI implementation. The former has more knowledge of drug discovery, development, and manufacturing, while the latter brings broader knowledge and experience to pharma projects, sharing insights from other contexts that have value for pharma applications.
It is also worth noting that some pharma companies choose to license AI technology that is implemented and managed by their in-house AI groups rather than forming deep relationships with tech companies. Others establish internal AI expertise through the acquisition of their AI partners. Still, others may outsource AI activities to a contract research organization that specializes in these activities.
It is not only strategies for the pursuit of AI projects that vary. Many different AI platforms have been adopted in the pharma industry to meet specific needs in different domains. In precision medicine, for instance, AI systems are used that can rapidly match specific therapies to individual patients. In manufacturing, meanwhile, AI platforms designed to select the best buffer solutions for the optimization of the entire production process are appropriate. Image screening and diagnostics applications require AI systems designed for rapid image processing. Platforms for use in clinical trials must be designed for comprehensive, compliant, and efficient clinical trial data management.
The need for specific AI solutions for different applications is perhaps one of the biggest challenges to leveraging the technology. Each system should be optimized for each application and project. Numerous platforms have been developed; unfortunately, sufficient data has yet to be generated to validate many of these systems. Validation is essential to ensure that generated results have real-world applicability and are part of the optimization process. Collecting data to support optimization and validation is therefore a current critical need.
Aside from the need for more data to populate AI systems, the technology itself is perhaps 80–90% ready for use in biopharma applications, with only moderate additional innovation in the technology itself needed to realize its potential. It is already revolutionizing fields such as physics and chemistry, and, with certain conventional theories, AI has actually surpassed human capabilities. The next step in the pharma sector will be feeding more data into the different systems in order to optimize and validate them. Going forward, we anticipate the continued improvement of systems targeting different pharma domains.
Looking at the industry as a whole, at least 60–70% of biopharma companies are bringing some sort of AI into their development programs. That includes smaller companies and startups, as well as Big Pharma firms. Some are leveraging AI to analyze phenotypic and genotypic data for target identification, while others are applying AI for drug screening to identify candidate molecules. As mentioned previously, it is also being widely applied in image analysis for diagnostic applications. Smaller firms with limited resources particularly benefit from the reduced time and cost that AI affords.
During the past 30 years, two to three different waves of AI implementation have occurred in different industrial sectors. In the pharma industry, the past few years have experienced the greatest acceleration of AI adoption, largely to address the unique situations created by the COVID-19 pandemic. The need to rapidly develop vaccines against the SARS-CoV-2 virus, for instance, led many to consider how AI can accelerate drug discovery, development, and manufacturing to benefit mankind.
In parallel, AI is seeing expanding use in other industries, with software, computers, and robotics in the lead. Robotics is potentially the largest opportunity at present, considering the ongoing development of self-driving cars and other similar technologies. As advances are made in these areas outside biopharma, an understanding of new ways in which AI can be applied and provide benefits is emerging, and this new knowledge will ultimately be leveraged for drug discovery, development, and manufacturing as well, driving future application waves in the pharmaceutical industry.
Sino Biological offers the world’s largest selection of bioactive recombinant proteins (nearly 6,500) and antibodies (monoclonal, polyclonal, and multispecific), as well as custom manufacturing and research services. AI is leveraged for drug discovery through a partnership with Ainnocence,a which has developed SentinusAI™ and CarbonAI™, self-evolving AI platforms specifically for the acceleration of large molecule and small molecule drug discovery, respectively.
Sino Biological and Ainnocence have joined forces to establish an AI-enabled platform for antibody affinity maturation. Powered by a self-evolving AI engine, Ainnocence's SentinusAI™ is revolutionizing affinity maturation processes. SentinusAI™ can effectively rank up to 1010 antibody sequences based on their predicted affinity toward one or more antigens within a few days. Sino Biological then performs physical screening of only the top candidates, fully expressing the selected antibody sequences and performing affinity validation studies based on its high-throughput platform for recombinant antibody development and advanced technologies for biomolecular interaction analysis.
The closed-loop nature of SentinusAI™ guarantees that the model creates better antibodies by learning from each cycle of the experimental results. A higher hit rate will be achieved on subsequent computation by incorporating wet-lab data. The AI-powered affinity maturation platform can deliver affinity-matured antibody sequences with an average hit rate of 15% and develop 103 increased affinity-matured antibodies within four weeks.
The Ainnocence AI platform is attractive because it only requires sequence data; no structural information must be provided. This approach dramatically reduces wet lab work, saving significant time and money. Through the partnership with Ainnocence and by leveraging its AI technology, Sino Biological has been able to further enhance its antibody development CRO services offering, thus saving customers precious development time and ensuring antibody–antigen binding affinities that meet their strict demands.
As the global leader in recombinant technology, Sino Biological has extensive experience in producing recombinant proteins and antibodies for research and drug discovery needs and can provide one-stop custom antibody services covering the initial antigen design to final scale-up antibody production. Sino Biological’s proprietary expression platforms and methodologies, as well as high-throughput and scale-up capabilities, ensure the highest chances of success for producing proteins and antibodies that are difficult to develop. Working with Ainnocence to offer next-generation antibody design and development CRO services, Sino Biological now has high expectations that AI can be leveraged to develop antibodies that will be effective in actual, real-life scenarios, even for applications in which the target is complex and/or continuously changing.
a Acknowledgment: The author acknowledges Dr. Lurong Pan, CEO of Ainnocence, for her invaluable input and contribution to this article.
Sumana Sundaramurthy is a Technical Account Manager at Sino Biological, where she supports R&D projects for industry and academic clients. Prior to joining Sino Biological, Sumana completed her doctoral studies in cell and developmental biology at SUNY Upstate Medical University. In addition to her academic pursuits, Sumana held various leadership positions supporting student life at Upstate and other professional organizations. She has also worked for a preclinical CRO and a few notable research groups at Loyola University, the University of Chicago, and Sanofi Pasteur.