Role of Generative AI in Drug Discovery

Generative AI-based LLMs have revolutionized the operations of various industries worldwide. Applications of Generative AI extend beyond LLMs and into more complex fields like healthcare, clinical development, and drug discovery. In this article, we focus on the role of Generative AI in drug discovery.

If you wish to learn Generative AI, consider enrolling in an Applied Gen AI course that covers all basic and advanced Gen AI concepts.

Subsequently, for a deeper understanding of Gen AI concepts like diffusion models, multimodal models, and reinforcement learning opt for this Advanced Gen AI course.

Also read the success stories of aspirants who have transitioned to Generative AI career.‍

Potential Of Generative AI In Drug Discovery

First, lets explore a few examples of the correlation between the languages we speak and human anatomy.

Human DNA can be viewed as a 3-billion-letter sequence forming a unique language of it’s own.

Similarly, as we can reuse 26 English language alphabets to create words and phrases, naturally evolved proteins can reuse their own alphabet comprising 20 amino acids. These amino acids can be rearranged and reassembled in innumerable ways to produce an infin of proteins, each with unique properties and roles.

By applying same methods used in LLMs, Generative AI can interpret these languages to uncover insights which used to go unnoticed with traditional methods. These capabilities of Generative AI can expediate drug discovery and reduce associated costs.

Thus, Generative AI has the potential to significantly put a curb on ever-increasing drug costs. 

Gen AI can help in every phase of drug development from its discovery to post market release surveillance.

According to various studies, with Generative AI, time required to discover and develop a drug gets reduced from ten years to one years. Cost incurred to gets reduced by nearly 70%.

Only 10% of discovered drugs reach clinical trails. Gen AI has the potential to reduce this failure rate and increase efficiency.

Benefits of Generative AI In Drug Discovery

As compared to traditional drug discovery methods, Gen AI has various benefits:

• Process: As compared to sequential approach followed in traditional drug delivery methods, Gen AI makes the process iterative.
• Effort: With Gen AI, the effort gets automated as machine learning algorithms generate drug molecules, compose trial protocols, and predict success during clinical trials.
• Time and Cost: Gen AI drug discovery methods are far less time-consuming and cost as little as one-tenth compared to traditional methods.
• Data Used: In traditional methods, data used is limited to current drug discovery experiments. However, with Gen AI data expands to extensive data sets of genomics, clinical compounds, scientific literature, and much more.
• Selection of targets: With Gen AI, the process can select several alternative targets as compared to only predetermined targets in traditional drug discovery methods.
• Personalization: Is limited to traditional drug discovery methods, however with the use of Gen AI, tailored drug candidates can be accommodated.

Also read: AI in Healthcare: Benefits, Examples, and more‍

Applications of Generative AI in Drug Discovery

Now let’s look at the various applications of Gen AI in drug discovery:

Molecule and Compound Generation

Train Gen AI algorithms on 3D shapes of molecules and their characteristics to produce molecules of desired properties. For instance, Insilicoused generative AI to generate ISM6331, a molecule with potential to target advanced solid tumors.

Train models to predict interactions between chemical compounds and recommend changes to balance their properties to mitigate risks.

Produce large sets of virtual compounds and evaluate them to find the best fit. For instance, Adaptyv Bio, uses generative AI algorithms to produce proteins.‍

Biomarker Identification

With Generative AI, algorithms can be trained to study large amounts of datasets to identify biomarkers against potential diseases. These biomarkers are then used to identify disease patterns in MRIs, CT scans, etc.

Since a large amount of datasets are analyzed with automation in considerably less time, the findings are more concrete and reliable.
Insilico Medicine built PandaOmics, a target identification tool that used Gen AI to identify biomarkers associated with gallbladder cancer and androgenic alopecia,

Also read: ML in Healthcare: Transforming Diagnostics and Treatment‍

Interaction Prediction of Drug and Targets

Gen AI models train with large datasets of biomedical data to predict the binding affinity and protein targets of new drug compounds.

These target proteins are then compared against innumerable chemical compounds to find any molecule capable of binding with the target.

Researchers at MIT and Tufts University used ConPlex, an LLM to evaluate drug-target interactions on 4,700 candidate molecules to test their binding energy. It found that 12 of the candidate molecules demonstrated very strong binding potential.

Repurpose and Combine Drugs

Gen AI algorithms can be used to test new therapeutic uses on existing drugs. Reusing existing drugs is much safer as they have a tested profile not present in untested drugs.

In addition to a single drug, Gen AI algorithms can repurpose the tests on various drug combinations to predict which one is best suited for a specific disorder.

Researchers employed Gen AI to find drug candidates for Alzheimer’s disease. They identified various drug candidates on 10 patients over 65 years of age, Their findings identified metformin, losartan, and simvastatin as drugs that demonstrated lower Alzheimer’s risks.

Researchers at IBM used a Gen AI algorithm to repurpose Rasagline, an existing Parkinson’s medication, and Zolpidem, used to ease insomnia. ‍

Also read: Transfer Learning: Leveraging Pretrained Models for Faster ML Development

Predict Side Effects of Drugs

One common problem with many drugs is the associated side effects. Gen AI algorithms can quickly aggregate data and simulate interactions with molecules to predict side effects that medical practitioners can use to identify the safest options.

Gen AI algorithms can predict precisely which biological processes in living organisms are affected by a drug molecule.

Gen AI can identify how different genetic makeups can respond to the candidate drugs.

The algorithms can study the correlation between drugs and adverse events to forecast side effects associated with these drugs.

They can also detect toxicity by binding non-target proteins and subsequently analyze drug-protein interactions to predict associated consequences.

In another drug discovery case, scientists from Stanford and McMaster University used Generative AI to produce molecules that have the capabilities to fight Acinetobacter baumannii. This discovery has the potential to deal with meningitis and pneumonia as Acinetobacter baumannii causes these diseases.

The Gen AI model was trained with a database of 132,000 molecule fragments and 13 chemical reactions to produce billions of candidates.

Another AI model screened these candidates against associated side effects to narrow the search down to six.

Challenges Related to Use of Gen AI for Drug Discovery

• Gen AI models lack reasoning capability: Gen AI models are like black boxes. They don’t provide any reasoning about how they work. But the scientists need to know the reasoning behind any occurrence.
  For instance, a model may inform about the side effects of a drug but doesn’t provide the reasoning behind the side effects.
• Gen AI models can produce inaccurate results: Large language models like ChatGPT produce results that are not 100% accurate. It translates to Gen AI model producing molecular structures that can’t be replicated in real life. But results generated by these AI algorithms, for instance a drug has no side effects, can be inaccurate as well.
• Gen AI models can be trained on incomplete data: If the dataset used to train a Gen AI model has incomplete information or information that is full of bias, then the algorithm would keep on producing results based on that incomplete information.
• Gen AI models struggle to explore chemical compound space: General purpose Gen AI models lack the capabilities to understand the vastness of chemical compound space. Some of them even resort to shortcuts like analyzing a 2D structure when 3D structures should be analyzed.
• Gen Ai models come with high infra costs: Building a Gen AI model from scratch is expensive. Even customizing a moderately large Gen AI model can incur expenses close to $200,000 with additional maintenance costs.

DeepMind AlphaFold

DeepMind AlphaFold is a Gen AI program developed by DeepMind, a subsidiary of Alphabet. Its latest version AlphaFold 3 was released on May 8, 2024. ‍

AlphaFold 3 can accurately predict the structure of proteins, DNA, RNA, selected ligands, and ions; and how they interact with each other. The model is reported to be trained with 170,000 proteins utilized from a public repository of sequences and structures.

AlphaFold 3 works on ‘Pairformer’, a deep learning architecture module.  Many predict that AlphaFold 3 has the potential to transform our understanding of the biological world and drug discovery.‍

Learn AI With Interview Kickstart

Interview Kickstart is a global leader in career upleveling. The edtech company offers a range of courses for candidates seeking a career switch to a lucrative and rewarding AI domain.

Interview Kickstart’s machine learning course covers every aspect of machine learning from foundational concepts to advanced machine learning frameworks. With capstone projects you also get hands-on experience of working with real-world applications.

IK’s data science course provides in-depth knowledge of concepts like data wrangling, machine learning, deep learning, data visualization, and big data analysis.

Our MLOps Training course is also highly appreciated as it covers every aspect of the MLOps lifecycle in detail.

In addition, IK offers many tech interview prep courses for aspirants planning to pursue a career in AI.

All courses are curated and taught by instructors who are tech leads and hiring managers at top-tier companies in the world. ‍

FAQs: Role of Gen AI in Drug Discovery

What Is the First AI Generated Drug?
The world’s first AI Generated Drug is INS018-055. Insilico Medicine, a biotech company headquartered in Hong Kong, created the drug which is also the world’s first AI-designed anti-fibrotic small molecule inhibitor drug. ‍

Give Some Examples of AI Generated Drugs?
A few examples of AI drugs are anastrozole, letrozole, and exemestane. They are also called aromatase inhibitors.  ‍

Who Introduced AI in Medicine?
Saul Amarel, a professor of computer science at Rutgers University who unveiled the Research Resource on Computers in Biomedicine in 1971 at the University, introduced AI in medicine.‍

Can AI Replace Doctors?
According to medical experts. AI can never replace doctors. However, the technology has the potential to assist medical practitioners.

Related Articles:

• ‍Generative AI vs Predictive AI: Everything You Need to Know‍
• How Companies are Implementing Generative AI? An Insider’s Look‍
• Revolutionizing Data Generation with Generative AI: Opportunities for Data Engineers”‍
• Mastering Generative AI: Your Roadmap to Getting Started‍
• Apple Finally Embraces Generative AI, So Should You!