Science Spyglass
How artificial intelligence is supporting the small molecule drug discovery process
Faster and more efficient
The discovery of new small molecule drug candidates has always been a complex, time-consuming, and costly endeavor. For decades, scientists have worked tirelessly to identify new chemical compounds that interact with specific biomolecules in the human body in therapeutically meaningful ways. Today, artificial intelligence in drug discovery is driving a major transformation of the process, providing scientists with powerful new tools to accelerate their work.
In simple terms, artificial intelligence (AI) refers to machines or computer programs that can perform tasks that usually require human intelligence, such as recognizing patterns, learning from experience, or making decisions. Machine learning (ML) is a specific type of AI that uses data to learn and improve over time without being explicitly programmed for every scenario.
Let’s explore how AI and ML are revolutionizing small molecule drug discovery and why it matters.
Enter artificial intelligence: smarter and faster drug discovery
This is where AI and ML come in. These technologies can process vast amounts of data quickly and identify patterns that humans might miss. Artificial intelligence in drug discovery can perform a wide range of tasks, including:
- Identify novel therapeutic targets: AI can integrate vast biological datasets to uncover new disease-related targets that may be overlooked by traditional methods.
- Predict the molecular architecture of protein receptors: AI predicts protein structure directly from the amino acid sequence, bypassing time-consuming and costly experiments.
- Predict how molecules behave: AI models estimate how a particular molecule might interact with a protein in the body, helping researchers decide whether it’s worth testing.
- Design new molecules: ML can generate entirely new chemical structures that have the potential to become effective drugs.
- Prioritize candidates for synthesis: AI and ML models assess ADME (absorption, distribution, metabolism, and excretion) properties, toxicity, and stability to help select the most promising compounds for experimental validation.
- Improve efficiency: By automating early steps in drug discovery, AI reduces the time and cost to bring a new drug to market.
Let’s break down each of these contributions.
1. Identifying novel targets
AI plays an increasingly important role in target identification, the critical first step in the drug discovery process. By integrating and analyzing large-scale omics datasets, including genomic, transcriptomic, proteomic, and clinical data, AI systems can detect complex biological patterns that point to previously unrecognized disease drivers. These insights can reveal novel, high-potential therapeutic targets and enable researchers to stratify diseases more precisely.
This data-driven approach goes beyond what traditional target discovery methods can achieve, particularly in areas like oncology, rare diseases, and multifactorial conditions. AI models can also prioritize targets based on predicted druggability and relevance to patient subgroups, ultimately helping de-risk early discovery efforts.
2. Predicting protein (receptor) structure
AI helps predict protein structures by analyzing amino acid sequences and determining how they fold into a unique 3D shape. This structural information is essential, as a protein’s shape defines its role in the body and forms the receptor site where small molecule drugs can bind. These interactions rely on shape complementarity and non-covalent molecular forces within specific binding pockets.
AI-based systems used to predict protein structures are trained on thousands of known examples; one such system is AlphaFold. These systems identify folding patterns and structural motifs, enabling them to predict the 3D conformation of proteins from their amino acid sequences with remarkable speed and accuracy.
In essence, by acting like an intelligent puzzle solver, AI dramatically accelerates structural biology. The significance of this breakthrough was recognized by the Nobel Prize in Chemistry awarded to the developers of AlphaFold.
3. Predicting molecular behavior
At the heart of many diseases is a faulty protein. Scientists aim to find molecules that can bind to these proteins to inhibit or activate their function. The ability of small molecules to recognize and bind to specific pockets on the protein receptors relies on factors such as shape complementarity, molecular interaction forces, and thermodynamic considerations.
Traditionally, this challenge is addressed through High-Throughput Screening (HTS), where hundreds of thousands of compounds are tested experimentally using suitable bioassays, extensive compound libraries, and laboratory automation platforms. In contrast, by digitally simulating these molecular interactions, AI enables the virtual screening of vast compound libraries, including molecules that have not been synthesized yet. This dramatically expands the search space and can potentially save time and resources. At some point, however, predictions need to be confirmed by real experiments for which compounds need to be purchased and prepared for testing or potentially synthesized from scratch.
Importantly, HTS and AI-based virtual screening are not mutually exclusive approaches and can be used in a complementary fashion in modern hit discovery. Insights obtained from virtual screening can guide HTS efforts by focusing experimental resources on the most promising compounds. Conversely, HTS can uncover active compounds that might be missed by current models, providing valuable data that can be fed back into AI algorithms to improve future predictions. When combined strategically, AI and HTS create a powerful workflow that enhances both speed and success rates of identifying high-quality hits.
4. Designing new molecules from scratch
One of the most exciting applications of artificial intelligence in drug discovery is its ability to design new molecules. Generative AI acts as a virtual chemist, rapidly inventing novel synthesizable molecules and organizing them into virtual libraries to speed up drug discovery.
- Designing new molecules: AI models extract chemical rules and patterns from large datasets of known compounds and reactions. They use this knowledge to generate new molecular structures that align with specific goals, such as targeting a particular protein or exhibiting desired properties.
- Ensuring synthetic accessibility: Generative AI can predict whether a proposed molecule can realistically be synthesized using available chemical building blocks and known reactions. Some AI systems, like SynFormer, even generate the synthetic pathways alongside the molecules, ensuring that the designs are practical for real-world chemistry.
- Focusing libraries: Rather than generating random molecules, AI can build focused libraries tailored to specific drug targets (e.g. RNA, or protein-protein interactions) or desired chemical features, streamlining the search for promising drug candidates.
Scaling up: These AI-driven methods enable scientists to explore and construct virtual libraries containing millions to trillions of possible compounds, something unachivable through manual or traditional approaches.
5. Screening and prioritizing candidates
Once a list of promising molecules is generated, the next step is selecting which ones to synthesize and test in the lab.
ML models can predict key ADME properties, whether a molecule is likely to be toxic, and how stable it will be in the body. Knowing these properties helps researchers avoid unviable compounds early.
In essence, AI functions like an expert assistant rapidly scanning through enormous chemical libraries and flagging the most promising options, potentially saving scientists time and resources.
6. Integration of individual process steps in automated workflows
Once initial hits are identified, compound optimization by medicinal chemistry advances through iterative cycles of “design – make – test – and analyze”. This optimization process benefits from highly standardized assays for activity or property profiling, performed on automated systems that ensure high-quality data is generated efficiently.
Certain types of chemistry are particularly well-suited to high-throughput and automation. Some AI-driven companies (e.g. Excentia or Insilico Medicine) in the field claim that integrating compound synthesis and profiling in fully automated robotic workflow further accelerate the process speed and efficiency.
Challenges and limitations of artificial intelligence
Despite its power, AI is not a cure-all. There are still limitations to overcome:
- Data quality matters: AI systems are only as good as the data they are trained on. If the data is biased, incomplete or otherwise of low quality, the results can be misleading.
- Complex biology: Human biology is incredibly complex, and even the most advanced AI systems cannot fully predict how a drug will behave in the body. This limitation applies even to molecular interactions appearing simple at first glance.
Ultimately, the accuracy of AI model predictions depends heavily on the quality and depth of existing data and biological understanding.
Looking ahead
The future of drug discovery lies in collaboration between humans and machines. AI can handle the heavy lifting – analyzing data, generating ideas, and optimizing choices – while scientists bring the creativity, intuition, experience and critical thinking needed to steer the process effectively.
As AI technology continues to evolve, it holds promise for faster development of treatments for diseases that have long been difficult to target, including rare disorders, cancers, and neurological conditions like Alzheimer’s. For patients, this means new hope. The dream is a world where treatments are not only faster to develop but also more personalized, effective, and accessible.
Author’s reflection
Personal note from the author (yes, this piece has been written by a human, not AI – though AI helped in the revision process): The above may sound like another wave of technology promises we have heard over the last decades, all claiming to fix the inefficiencies of drug discovery process. In the end and all hype aside, AI is not a magic solution. What it does offer is a powerful extension of our scientific toolbox and contribute to tackle the increasingly challenging biological targets.
Related content - Accelerate your drug discovery pipeline
Learn how Axxam experimentally validates targets, pathways, and compounds identified through artificial intelligence