Digital Transformation: Accelerating Small-Molecule Drug Discovery

Developing a new drug can take more than 10 years and several hundred million dollars, with only a one in 30,000 chance of success. Accelerating the drug research and development process is essential for pharmaceutical companies to meet patient needs more quickly and efficiently.

Pharmaceutical companies are leaning on artificial intelligence (AI) as the enabling technology for unprecedented productivity (1). In an industry that handles vast amounts of data throughout its entire value chain—drug discovery (research), development, manufacturing, sales, and post-marketing maximization of product value—the impact of AI-driven drug discovery and development is expected to be significant. In fact, it has been estimated that the cost of R&D can be reduced by about 60% and the time period shortened by about two and a half years by advancing digital transformation (DX) (2).

DX is one of the critical enablers to achieving a company’s goals. In one example, Astellas has been driving forward DX initiatives throughout its value chain, placing emphasis on the best mix of people and digital technology. The company’s approach is to build the best possible relationship between people and digital technology to achieve what has been unattainable until now.

Combining the capabilities of AI and robots with the skills and experience of people, rather than using AI and robots as replacements for people, the pharmaceutical industry can engage in state-of-the-art drug discovery that enables the development of high-quality drugs in a shorter time. In the case of Astellas, a “Discovery Intelligence” team was formed to focus on developing and integrating intelligence technologies such as AI and automation robotics into the candidate creation processes by digitizing the company’s drug discovery platform. The company is now advancing research programs at greater speed than has been previously experienced.

Drug discovery digital transformation

The value chain of drug discovery research starts with the researcher’s idea. The key is to find the unmet medical need, identify the cause of the disease, and select the best modality for the therapeutic target.After the modality is selected with the right target molecule (protein, gene, etc.), compounds are created that match the therapeutic target.

With Astellas’ DX approach, its researchers have been able to shorten the drug discovery period, as shown in Figure 1. The upper panel shows the conventional approach, and the lower panel shows the shortened time period with DX. There are two major challenges to DX: obtaining a hit compound against a target molecule to find the lead compound and optimizing the lead compound into a candidate compound, which can take up to three and a half years. With the introduction of AI and robotics, these periods have been dramatically shorted in recent years—the company has seen an approximately 70% reduction from the total three and a half years in empirical time span within Astellas drug discovery—especially because of the diversity of library and implementation of digital technology.

Ultra large-scale virtual screening

At Astellas, the company has two points of differentiation for hit identification utilizing protein structure analysis. First, high-quality protein preparation and crystallization and structural analysis technologies enable rapid acquisition of unique structural information with the company’s accumulated compound-protein complex structure data (about 10,000 structures) for activity prediction. Second, the company’s drug-like virtual library of hundreds of millions of units generated from approximately 23,000 building blocks that have been scrutinized and collected by the company’s medicinal chemists via visual inspection enables rapid synthesis of compounds that are hits in ultra large-scale virtual screening (ULVS).

Compounds in the company’s protein structure and virtual chemical libraries are drug-like and can be rapidly synthesized from ULVS hit compounds into drug candidates for development. These clinical assets are supported by cloud computing to run large-scale computer simulation (Amazon Web Service). The ULVS platform, which is a mix of human and artificial intelligence, is depicted in Figure 2.

Human-in-the-loop drug discovery

By combining human expertise (e.g., medicinal chemistry knowledge) with AI and robotics (protein structure analysis technology, virtual library, and cloud computing), a company can shorten the timeframe of small-molecule discovery without sacrificing quality. In addition, in the case of Astellas, which has been able to cultivate a comprehensive database of small-molecule drug discovery targets throughout its company history, this approach of combining human experience and digital tools enables the company to find the seeds of drugs in a short period of time.

The human-in-the-loop (HITL) drug discovery platform (3) revolves around a design, make, test, and analysis (DMTA) cycle, as shown in Figure 3 and as described below:

• Design: by utilizing AI and robots together with researchers' input and ideas, AI-assisted chemical structure designwith experimental dataset, and comprehensive judgment, one can significantly speed up the drug discovery process. For example, the HITL drug discovery platform was able to reduce the time it took from hit compound to acquisition of a drug candidate compound by about 70% (from two-and-a-half years to seven months).
• Make: In this example, Astellas has historically incorporated high-throughput synthesis technology and has built a system for parallel synthesis of multiple samples by utilizing a 23,000-building block library and automated equipment that allows for rapid acquisition of a large amount of data. The company then accelerated automation of compound synthesis with the introduction of a robotic synthesizer (Chemspeed Technology AG) in 2021 and has now automated 14 different reactions and several work-up processes. Its goal is to automate 50% of commonly used reactions in the near future.
• Test: Astellas also developed and implemented an automation system that it nicknamed “Screening Station”—a flexible robotic system that enables high quality and uniformity in cell assay system screening. This automation system contributes multiple samplesfortesting efficacy. Through the accumulation of the efficiency gains described above, Screening Station enables experiments to be conducted that are 100 to 1000 times larger than previous research conducted in the same amount of time. The full automation of experiments has also created a more efficient workflow and reduced the possibility of human error. Researchers no longer need to do the routine work between experiments, such as changing plates, for example, which leaves more time for data analysis, future experiment planning, and the development of mid- to long-term strategies and plans.
• Analysis: Astellas’ compound database and machine learning platform (DataRobot) were linked to the company’s workflow platform (KNIME), enabling it to automate data extraction and preprocessing to prediction model construction. A noteworthy feature of this system is the automation of the structure-activity-relationship (SAR) table creation. The SAR table creator can quickly visualize various predicted and actual evaluation results together with compound structures to help chemists easily understand SARs.In addition, since the table can be used to create materials for presentations, there are secondary effects, such as a significant reduction in the time required for researchers to create materials and elimination of transcription errors.

Beyond small-molecule drug discovery: cell and gene therapy
Astellas aims to expand drug discovery utilizing DX to new modalities, such as biologics and cell and gene therapies. To accelerate research in cell therapy, the company developed its own DX-enabled drug discovery platform, the Mahol-A-Ba platform, the company’s newest HITL drug discovery approach. The platform leverages the Maholo LabDroid (Robotic Biology Institute)—a robot that had already been introduced at the company’s Tsukuba Research Center in Japan to utilize induced pluripotent stem cells (iPSCs) for drug discovery (4).

Similar to Screening Station, Mahol-A-Ba allows Astellas to conduct significantly larger experiments in the same amount of time and has brought forth more efficiencies, as well as lessened the likelihood of human error.

Bringing valuable medicines to patients

Throughout the R&D process, from bench to clinic to patients, taking a “science first” approach optimizes the chances of creating new treatment options and maximizes value for patients with high unmet needs. Taking a science first approach means focus should be concentrated on the best science, empowering the best talent to pursue that science, and developing it at a most conducive location.

The goal with DX is to create the best mix of human and digital resources. By advancing these initiatives, pharmaceutical companies such as Astellas can provide valuable medicines for patients and potentially define entirely new chapters in the treatment of disease.

References

1. M.K.P. Jayatunga, et al., Nature Reviews Drug Discovery 21, 175–176 (2022).
2. T. Nagakawa/Deloitte Tohmatsu Consulting, Paradigm of New Drug Discovery through Technology Advance (2018).
3. Astellas Pharma, “A Human-in-the-Loop Drug Discovery Platform Integrating Humans, AI, and Robots,” Video on YouTube.com, July 8, 2022.
4. M. Sasamata, et al., SLAS Technology 26 (5) 441–453 (2021).

About the Author

Kenji Tabata, PhD, is senior vice-president, head of Discovery Intelligence, Applied Research & Operations, at Japan-based pharmaceutical company Astellas.