The process of taking a drug molecule from concept to commercialization is a lengthy and costly endeavor, typically spanning 10-15 years and incurring expenses of up to $2 billion per successfully launched drug when factoring in failures. In this article, we delve into the potential of artificial intelligence (AI) and machine learning (ML) to expedite drug discovery, shedding light on their impact and limitations.
The need for accelerated drug discovery
The early discovery phase of drug development incurs significant costs and demands innovative technologies to expedite and de-risk the pipeline. AI and ML have emerged as key players, offering the capacity to broaden the chemical “search space” for novel compounds, streamline complex property calculations, and provide insights into often noisy and incomplete data.
The selection of the best candidate molecule for progression from discovery to a new chemical entity (NCE) is a complex and time-consuming task. With the growing difficulty of finding novel molecules to target current drug interests, researchers are eager to evaluate a greater number of compounds without exponentially increasing budgets for wet chemistry and biology.
AI and ML, running on cutting-edge hardware, are revolutionizing this landscape. They can triage molecules at an unprecedented scale, allowing the exploration of vast chemical libraries. AI-driven approaches range from constructing “virtual libraries” of molecules generated through synthetic reactions to using “generative AI” to assemble molecules from a molecular soup of small fragments. Each approach has its strengths and challenges, including issues related to molecule accessibility and diversity.
Accelerating drug discovery through AI/ML
AI for drug discovery (AIDD) is still in its early stages but has already yielded compounds entering clinical trials. The impact of AI and ML is multi-faceted:
1. Complex calculations made simple:AI/ML tools are trained to predict results of intricate calculations that determine a drug candidate’s behavior. For instance, AI can approximate the quantum mechanics of compound libraries, dramatically speeding up the analysis of electronic contributions to a molecule’s properties.
2. Binding Energy Predictions: AI can predict binding energies through free energy perturbation (FEP) calculations. This approach, traditionally resource-intensive, now benefits from AI methods, which accurately screen libraries with FEP calculations, yielding results comparable to experiments.
3. Protein structure prediction: AI models can generate protein structures for targets that lack experimental data. While this field has faced challenges, the sophistication of AI tools and understanding of their output are growing, making AI-generated protein structures increasingly relevant in drug discovery.
4. Streamlining screening: AI models can efficiently screen ultra-large libraries of molecules against specific drug targets. By rapidly evaluating billions of potential drug candidates, AI/ML streamlines candidate analysis, allowing researchers to focus on the most promising compounds.
5. Synthetic route optimization: AI excels at pattern recognition in complex datasets, making it proficient in generating and optimizing synthetic routes to desired molecules. AI’s ability to find efficient routes in less time than experienced chemists is a significant advantage.
Overcoming limitations for wider AI adoption
AI/ML’s impact on drug discovery is substantial, but challenges remain. AI models require extensive datasets for training, and data sharing in the pharmaceutical industry is restricted due to privacy and confidentiality concerns. To fully unlock AI’s potential, the industry must become more open to data sharing, allowing AI/ML tools to predict complex properties accurately.
Another challenge is the scarcity of experts proficient in both drug discovery and AI, hindering efficient tool utilization and insights generation. Addressing these challenges will be pivotal in maximizing AI’s role in drug discovery.
The future of AI in drug discovery
While AI is unlikely to replace all aspects of drug discovery, its potential for positive impact is evident. Ongoing developments in quantum mechanics, synthetic route optimization, protein structure prediction, and other areas will continue to shape AI’s role in drug discovery. As more data becomes available, AI/ML tools will be better equipped to address complex calculations.
The integration of AI/ML into drug discovery platforms has already led to significant improvements, enhancing efficiency and reducing costs. These technologies will play an ever-growing role in the pharmaceutical industry, accelerating candidate selection and transforming the landscape of drug development.