AIDD: When Artificial Intelligence Begins to Design Drugs

From the 2024 Nobel Prize in Chemistry to the growing role of AI in drug discovery.

When AI Enters the Molecular World

What if we could design proteins the way engineers design airplanes, rather than simply studying the molecules that already exist in nature?

This is no longer just a science-fiction concept.

The 2024 Nobel Prize in Chemistry recognized groundbreaking advances in protein science. Demis Hassabis and John Jumper were honored for developing AlphaFold2, an AI system capable of predicting three-dimensional protein structures from amino acid sequences. David Baker received the prize for his pioneering contributions to computational protein design, demonstrating how computational methods can be used to create proteins with novel structures.

Together, these achievements illustrate how computational technologies are helping scientists move beyond understanding protein structures toward designing new proteins with desired characteristics.

A closely related transformation is also taking place in pharmaceutical research: AI-Driven Drug Discovery (AIDD).

What Is AIDD?

Traditional drug discovery often involves screening, testing, and optimizing large numbers of candidate molecules to identify compounds with promising therapeutic potential and acceptable safety profiles.

AIDD introduces a new set of tools into this process. By learning from extensive biological, chemical, and experimental datasets, AI can help researchers identify potential drug targets, generate candidate molecules, predict molecular properties, and optimize existing designs.

Think of AI as a research assistant capable of analyzing enormous amounts of information and identifying patterns that might otherwise be difficult to recognize.

It cannot reliably determine which molecule will ultimately become a successful drug. However, it can help researchers explore molecular possibilities more efficiently and prioritize candidates for experimental investigation.

The value of AIDD is not to eliminate experiments, but to make experimental research more informed and targeted.

Can AI Really Design Antibodies?

Antibodies are an important application area for AIDD.

Antibodies are proteins that recognize specific molecular targets, or antigens. Their interactions with antigens can be loosely compared to keys fitting into locks. Traditional antibody discovery often involves screening large collections of antibody candidates to identify promising binders, followed by experimental optimization.

AI offers another possibility: What if we could use information about the lock to help design a better-fitting key?

By analyzing antibody sequences, three-dimensional structures, and antibody–antigen interaction data, AI models can propose new candidate sequences and predict potential binding characteristics. This allows researchers to explore vast sequence spaces and prioritize candidates for experimental validation.

AI can also assist in optimizing previously identified antibodies. For example, affinity maturation involves modifying antibody sequences to improve their binding interactions with an antigen. AI-based approaches can help identify potentially beneficial amino acid substitutions and guide experimental optimization.

Another application is antibody humanization, in which nonhuman antibody sequences are modified to more closely resemble human antibodies while aiming to preserve their binding properties. Computational approaches can support this process and help assess potential immunogenicity-related sequence features.

Beyond target binding, successful antibody development also depends on molecular properties that influence manufacturing and downstream applications. AI can help predict characteristics such as stability, solubility, aggregation propensity, and expression-related behavior, supporting early developability assessment.

From candidate generation and molecular optimization to developability prediction, AI is expanding the possibilities of antibody engineering.

However, a key that appears perfect in a computer simulation may not open a real lock.

Likewise, an AI-designed antibody may exhibit poor expression, insufficient stability, or unexpected functional behavior. Experimental expression, purification, and functional characterization remain essential for determining whether a candidate performs as intended.

AI and Experimental Biology: A Continuous Learning Cycle

One of the most promising aspects of AIDD is the opportunity to connect computational predictions with experimental research through iterative feedback.

AI models propose molecular designs based on available information. Laboratory experiments reveal how these candidates actually perform, while newly generated data can inform subsequent rounds of prediction and optimization.

This approach transforms drug discovery from a process of repeatedly testing candidates into a more integrated cycle of knowledge generation and data-driven decision-making.

The 2024 Nobel Prize in Chemistry did not mark the beginning of AIDD. Instead, it recognized foundational scientific advances that demonstrate how computational approaches are reshaping our ability to understand and design biological molecules.

AI will not turn drug discovery into a one-click process, but it is already changing how scientists search for the next generation of medicines.