Since AI’s effectiveness relies on the availability of high-quality, comprehensive data, the detailed drug discovery tasks will evolve to generate superior-quality data automatically.
Hanjo, what initially sparked your interest in the integration of AI/ML with the fields of organic chemistry and drug discovery science?
With a lengthy background in organic chemistry, molecular modeling, and cheminformatics, I initially perceived AI as a mere synonym for the familiar technology, Computer-Aided Drug Design (CADD). However, two pivotal experiences led me to pivot my career, transitioning from a pharmaceutical company in Korea to Standigm.
The first catalyst for my shift was a drug discovery project I was leading. We were grappling with issues related to our lead compound’s scaffolds and sought the assistance of an AI-powered drug discovery startup to propose new structures. My initial skepticism towards their suggested compound list was proven wrong when one of their compounds emerged as the most effective in vivo hit compound we had ever synthesized. This revelation made me reflect on the potential bias in my previous work, which may have overlooked numerous promising compounds. Unlike me, the AI algorithm, albeit in its early stages, had no bias and offered practical solutions.
The second reason relates to the role of generative chemistry models in the Design-Make-Test- Analyze (DMTA) cycle. Traditional CADD technologies such as QSAR and virtual screening are instrumental in proposing potential hit or lead compounds during the initial stages of drug discovery. However, these technologies merely guide medicinal chemists in designing novel molecular structures and selecting some for synthesis. The process of designing new molecular structures based on data analysis calls for highly skilled scientists with extensive and varied experience, making drug discovery an incredibly time-consuming and resource-intensive task. While not as adept as human experts, generative chemistry models offer structure design capabilities, facilitating automation in the DMTA cycle and leading to model improvements.
These two experiences convinced me of the value of AI in drug discovery, leading to my decision to join Standigm in 2019.”
How does Standigm utilize cutting-edge algorithms and computational methodologies to transform and optimize the drug discovery process, and what specific types of algorithms are integral to their innovative approach?
Two of the most crucial tasks in drug discovery encompass target identification and lead compound creation. Standigm’s technology platforms, Standigm ASK and Standigm BEST, are specifically engineered to tackle these key challenges.
Standigm ASK comprises an array of algorithms designed to identify promising and novel target proteins for a disease. This platform utilizes a dual approach: firstly, it employs a knowledge- level approach, leveraging various AI algorithms to reveal hidden relationships between target proteins and diseases within an extensive biology knowledge graph database. Secondly, it applies an experimental data approach, utilizing a multitude of omics data from real-world samples to capture the biological contexts of systems of interest, be it a cell, tissue, organ, or living organism. The amalgamation of these two unique approaches facilitates translational research, linking basic science results with real-world data.
Similarly, Standigm BEST is a repertoire of various generative and predictive models aiming to identify hit compounds, design innovative scaffolds during hit-to-lead, and optimize lead molecules to enhance activity and ensure appropriate druggability.
Hanjo, can you tell us about your professional background and your current role at Standigm. Also tell us how does Standigm differentiate itself from other companies in the same space?
Since earning my Ph.D. in organic chemistry in 2004, my career has spanned various research institutes, startups, and pharmaceutical companies, where I have focused on molecular modeling and cheminformatics.
At Standigm, I spearheaded the development of Standigm BEST, leading a multidisciplinary team that included medicinal chemists, computational chemists, software developers, AI scientists, and biologists. In 2022, my role expanded to encompass business development, management of our overseas offices in Cambridge, UK (Standigm UK), and Boston, US (Standigm US), as well as public relations activities. With the appointment of our new CEO, Dr. Youngsung Choo, in April 2023, my focus shifted from research to strategic planning for the company.
Standigm maintains an optimal balance among three expertise areas: chemistry, biology, and AI science. This equilibrium is critical for AI drug discovery companies to sustain balanced operations. We collaborate with complementary partners to forge the best technology and data integration for drug discovery. Standigm ASK differentiates itself from competitors through its transparent algorithms, enabling human experts to formulate hypotheses for validation.
Standigm BEST employs a unique method of representing the electrostatic potential of small molecules in a fingerprint format, resulting in superior predictive models, even when using the same AI algorithm.
How has AI/ML technology impacted traditional drug discovery sciences, and what challenges and opportunities arise from the intersection of AI/ML and drug discovery?
Generative chemistry models, which I’ve previously discussed, can be applied across any AI/ML technology. They offer novel ways of integrating distinct tasks in drug discovery projects, allowing human experts to concentrate on higher-level decision-making without getting mired in repetitive and tedious tasks, thanks to the adoption of AI/ML-based modules.
Expertise in AI/ML and drug discovery share few commonalities, with differences spanning education, philosophy, working methodology, and terminology. Drug discovery projects necessitate a diverse range of experts for continuous decision-making. This can lead to complex dynamics, underscoring the importance of facilitators or translators in this context. Bridging these two distinct worlds is a challenging yet rewarding endeavor.
How do you see the future of drug discovery evolving with the continued integration of AI technologies, and what potential impact could it have on the pharmaceutical industry?
Incorporating AI into traditional research environments enhances capabilities and efficiency, though it demands adaptability. Since AI’s effectiveness relies on the availability of high-quality, comprehensive data, the detailed drug discovery tasks will evolve to generate superior-quality data automatically. This change, coupled with the necessity for data integrity even in early-stage research, will expedite higher-level automation, ensuring unexpected human interventions don’t disrupt the entire data production process.
Many envision the future of the pharmaceutical industry as 4P medicine: predictive, preventive, personalized, and participatory. The industry is likely to adopt AI/ML technologies more assertively to realize this vision. The transformation of pharmaceutical factories may mirror that of the silicon industry, with the development of smarter factories capable of producing small quantities of various drugs — potentially even individualized medicines.
For participatory and preventive medicine, big tech companies’ expertise in acquiring, managing, and analyzing personal health data must intersect with drug discovery and development technology while maintaining data privacy. The collaboration between pharmaceutical companies, big tech firms, and technology-providing startups will be instrumental in disrupting the future landscape of healthcare and life sciences.”
How does natural language processing contribute to the extraction of valuable information from scientific literature to enhance drug discovery processes?
Standigm ASK incorporates a Natural Language Processing (NLP) module that translates journal publication abstracts into graph network databases. This module ensures that the knowledge database remains current. Given the capabilities of Large Language Models (LLMs), NLP is expected to evolve into LLM to comprehend scientific contexts better, resulting in a more extensive, contextually aligned database. A significant aspect of scientific data for drug discovery and development is the constant insufficiency of data to feed AI models like LLMs.
Deeper scientific exploration demands more precise contexts, restricting the quantity and quality of available data. At the forefront of any scientific field, high-quality data is always limited.
The task of capturing Structure-Activity Relationship (SAR) data largely falls on companies in India due to their cost-effective solutions. The situation is further complicated by the reluctance of scientific papers or patent authors to provide machine-readable data in a standardized
format while maintaining data quality. For instance, organic reaction data is infamous for low reproducibility of experiments detailed in even highly-respected journals, as extracting all essential information from texts presents a challenge. Utilizing LLM can at least eliminate inconsistencies in the data-capturing process by different individuals, potentially leading to a higher-quality database suitable for AI model learning.
Foundational models that comprehend scientific areas hold the potential to revolutionize how we understand and use highly specific data in broader contexts.
What advice do you have for other leaders who are looking to drive growth for their brand?
While I wouldn’t claim to have achieved significant success in my field, I believe that an open and humble mindset is critical in multidisciplinary sciences, such as AI-driven drug discovery. This mindset allows one to traverse various fields of expertise and innovate in our work methodology. The most compelling draw for employees is working with accomplished individuals who inspire them to improve and advance. A cross-functional team, whose members are open to different viewpoints and willing to adapt their methods in line with their shared goals, can achieve those goals effectively. Business growth naturally follows this process.
What is the biggest problem you or your team is solving this year?
As an AI-driven drug discovery company, Standigm emphasizes the balance of expertise, as previously mentioned. However, due to limited resources, this also implies potential shortfalls in other capabilities necessary for drug discovery projects, such as clinical study expertise. Market expectations often pressure startups to demonstrate promising results to gain credibility, with clinical-stage assets becoming a standard requirement for AI drug discovery companies like Standigm.
Standigm’s primary challenge is generating clinical-stage assets despite limited expertise, prompting the appointment of a new CEO with the necessary experience.
A consistent issue, not specific to this year, is how Standigm can become part of the dynamic ecosystem of the pharmaceutical industry. Overcoming obstacles such as language barriers, geographical location, and human resources remain key challenges.
Is there anything that you’re currently reading, or any favorite books, that you’d Recommend?
The most recent book I indulged in a few weeks ago was “Creative Selection” by Ken Kocienda, a former Apple engineer. The book sheds light on the single culture behind Apple’s astounding successes and how ‘creative selection’ serves as an atomic process that can enhance systems, companies, people, and projects — a concept that resonates deeply with the workflows we’ve established for AI-driven drug discovery. While I can’t predict the success of our endeavors, I find them far from daunting, primarily because of my passion for them.
Thanks Hanjo!
Hanjo Kim is the SVP of Global Strategy, Head of Medicinal Chemistry at Standigm, a workflow AI-driven drug discovery company headquartered in Seoul, South Korea with offices in the U.S. and UK. As SVP of Global Strategy, he leads the company’s business development and public relations teams and as Head of Medicinal Chemistry, he leads the medicinal chemist team and the AI scientist team, focusing on organic synthesis-related AI models. This combination of two groups enables the generative models of Standigm to incorporate the patentability, synthetic accessibility, and activity of molecules in its design process. Prior to joining Standigm, he worked as the Principal Researcher, CJ Healthcare R&D Center, Head, EQUIS & ZAROO R&D Center, and Principal Scientist, EQUIS & ZAROO R&D Center. He holds a Ph.D. in organic chemistry from Yonsei University.
Standigm is a workflow AI-driven drug discovery company. Standigm has proprietary AI platforms encompassing novel target identification to lead generation to generate commercially valuable drug pipelines. Founded in 2015, the company has established an early-stage drug discovery workflow AI to generate First-in-Class lead compounds within seven months. Pursuing full-stack, AI-driven industrializing drug discovery, Standigm has achieved the automation of molecular design workflow through DarkMolFactory™, and the automation effort has been expanding to the whole drug discovery process on the basis of Standigm AI platforms, including Standigm ASK™ for target discovery, Standigm BEST™ for lead design, and Standigm Insight™ for drug repurposing.