Generative AI and Modern Drug Development

To get our insights in your email, Subscribe to our Newsletter.

How long does it take for life-saving drugs to be invented? Before 2019, the fastest vaccine that had ever been developed was the mumps MMR vaccine, which took four years. It can take as much as fifteen years for vaccines to be developed. In 2019-20, of course, something that had never happened before took place. A vaccine was developed for the SARS-CoV-2 virus in less than a year.

How was the COVID-19 vaccine developed so quickly? One major reason is that Moderna had installed AI systems many years earlier. The AI systems were developed to accelerate research and drug development. And it worked! In less than two months, the vaccine had been made ready for human trial.

The COVID-19 vaccine was a rare case of urgent international collaboration in the face of a global emergency. Can the development life cycle of life-saving medicine be truncated in the absence of a worldwide pandemic?

The future lies in Generative AI for drug development.

Generative AI is not just a buzzword; it’s a powerful tool transforming drug development. Drug development has always been a long, costly process. Generative AI is set to make drug development faster, cheaper, and more efficient. Here’s how.Understanding Generative AIThe purpose of Generative AI is to create new content. This could be anything from text and images to molecular structures. Advanced algorithms like Generative Adversarial Networks (GANs) and transformer models help in this way. Each has its roles and capabilities in the context of drug development. GANs and transformer models significantly enhance various stages of the development process.

Generative Adversarial Networks

GANs consist of two neural networks: the generator and the discriminator. The generator creates synthetic data samples, which the discriminator evaluates against real data. Over time, the generator improves, producing increasingly realistic data.

This has applications in molecule generation and data augmentation. The generator creates new molecular structures that could serve as drug candidates. Their validity is assessed by the discriminator. Iterations are undertaken until the system identifies promising candidates.

These AI models can also create synthetic biological data. Such data can be used to augment limited datasets and improve AI model training.

Transformer Models

This deep learning AI model is primarily used in NLP (natural language processing). They’re built on layers of self-attention and feedforward neural networks. These mechanisms weigh the importance of different words in each sentence, letting the AI model better understand the text and context. Transformers can handle long-range data dependencies.

This means that the transformer can quickly mine complex text, like scientific literature. Despite the length and complexity of the content, the model can extract relevant information. It delivers highly specific answers and text output. Such information helps with understanding drug interactions, side effects, and other biomedical research findings.

Transformers also help with protein folding and sequence prediction. They deliver information that researchers can use to identify viable drug candidates. Researchers can better understand how the drug would interact with the target disease.

Generative AI Drug Development in the Real World

In 2023, Insilico Medicine became the first to bring a drug discovered and designed by generative AI into Phase II clinical trials. This is for a pan-fibrotic inhibitor, INS018_055, potentially the first of its kind. It was designed to treat idiopathic pulmonary fibrosis (IPF).

INS018_055 was developed using Insilico’s end-to-end AI drug discovery platform.

The drug development process began in mid-2019, with the preclinical candidate selected just 18 months later. Insilico announced Phase I human trials nine months later, less than thirty months from the novel target discovery. That’s less than half the time taken through traditional drug discovery.

The Benefits of AI-driven Drug Development

Accelerated Drug Discovery

Drug discovery is a complex, expensive process. It typically takes over a decade and billions of dollars to develop a new drug. This can be cut short by generative AI, making the process less expensive too. Thousands of molecular structures can be quickly generated and evaluated by the AI system. This helps researchers quickly identify the best candidates for further development.

Through virtual screening, AI/ML models can speed up drug candidate selection. In this process, the models parse vast datasets to identify promising compounds with the right structure and interactions. AI helped with the quick identification of treatments, especially during the pandemic.

For example, using AI tools like BenevolentAI, scientists were able to identify drugs that could also work as potential COVID-19 treatments. AI identified that the well-known anti-inflammatory, Baricitinib, had an off-target anti-viral effect - something that was validated during clinical trials. From start to finish, the entire process took just 48 hours.

Efficient Drug Development

The drug development process can be optimized by generative AI. The AI systems can also predict the most efficient ways to synthesize the selected compound. This further reduces costs and increases yield. AI analyzes patient data to improve clinical trial design. This is done to find the best participants and predict outcomes accurately.

Better clinical trials make development faster and more efficient. From drug discovery to design, the process usually takes 4-5 years. In just 12 months, Exscientia used AI to develop a drug molecule to treat OCD. AI can bring new drugs to market faster, benefiting patients and healthcare providers alike.

Precision Medicine and Personalized Treatment

Generative AI plays a key role in precision medicine. By analyzing patient data, it can develop personalized treatment plans tailored to individual genetic and medical profiles. Such highly targeted treatments can be extremely effective.

For example, Tempus uses AI to deliver precision medicine for cancer treatment. Their tools help doctors make data-driven decisions thanks to their analysis of clinical and molecular data. They reduce the risk of adverse reactions by identifying relevant biomarkers and drug targets. This further improves trial outcomes.

The Ethical Concerns

Generative AI in drug development delivers immense benefits. However, there are certain ethical considerations that need to be addressed. For example, if AI algorithms are trained on non-representative data, they can become biased. This can further lead to healthcare inequities. It’s possible to mitigate this issue by ensuring diversity in training datasets and monitoring AI outputs.

Another major concern is data privacy and security. AI algorithms ideally need to be trained on real-life patient data. This means that proper collection, maintenance, and security of this sensitive patient data are essential. Patient permission to use and process the data is also a question. To protect both healthcare providers and patients, regulatory frameworks need to evolve to keep up with AI advancements.

Challenges and Considerations

In addition to the ethical concerns, some challenges need to be addressed:

• Data Quality: The effectiveness of AI models depends on the quality and diversity of data they’re trained on. If the datasets are incomplete or biased, the improper data can result in ineffective or even harmful potential drugs.
• Processing Power: Training AI models requires significant process power. Established cloud-based AI tools may be more suitable.
• Investment: Integrating AI into existing development workflows demands investment in infrastructure and training. Resources of all kinds, including cost, expert manhours, and development time, are in high demand.
• Resistance: Biologists and drug researchers are often reluctant to adapt! While they may be willing to use the technology, developing the AI tools requires interdisciplinary collaboration between AI specialists and biologists, which is key to successfully leveraging generative AI in drug development.
• Interpretability: Understanding and interpreting the decisions made by these models can be complex. More transparent AI systems need to be developed.
• Intellectual Property and Regulatory Concerns: It’s hard to determine the ownership of AI-generated drug candidates, as AI-developed drugs are still in their infancy, and regulations are also still evolving.
• Validation: Biological systems are complex. It’s not yet safe to take an AI prediction at face value. Each AI prediction must be thoroughly validated through experimental and clinical trials.

Entering the Future: AI-powered Drug Development

Like so many other industries, generative AI is set to transform drug development. As the technology evolves, many of the concerns we’ve discussed above will fade away. Predictive modeling will become more accurate. Data selection (to train the model) will itself be AI-driven! Regulations and ownership rules will become established.

Collaboration between AI experts and drug developers will be vital to harnessing AI’s full potential in this technically challenging field. AI-driven biotech companies like Insilico will be at the forefront of this revolution.

From accelerated molecule design to optimized clinical trials and personalized patient care, AI’s applications in pharmacology are transformative. Generative AI is paving the way for innovative treatments.

How can generative AI benefit your organization? Ask us for a consultation today.

Follow us for more technology insights!