Artificial Intelligence Revolutionizes Drug Discovery and Development

Heather A Noriega, MS Pharmacology, Ph.D. candidate Pharmaceutical Science, Howard University College of Pharmacy

The pharmaceutical industry has long faced drug discovery and development challenges, from high costs to time-consuming processes that can take over a decade, with 90% of drug candidates failing during the development process. [1] Data from the Food and Drug Administration (FDA) has estimated that the cost of developing a new drug can range from $314 million to $2.8 billion. [2] The pharmaceutical industry is working to lower these costs by moving away from traditional ‘blockbuster' drugs, which are one-size-fits-all medicines that generate huge sales, to more specialized 'niche buster' drugs. [3] This combination of cost, risk, and time is a major barrier to innovation. Nevertheless, from 2018 to January 2024, the increase in projects and funding from the National Institutes of Health (NIH), ranging from $100-150 million, has made immense efforts, and the use of Artificial Intelligence (AI) in drug development is rapidly growing. [4] Researchers and pharmaceutical companies are now using AI technologies throughout the entire drug development process, from early research to after the drug is on the market. This has been shown to cut costs, reduce durations, and reduce risks. 

However, despite the advancements, AI has not come without other challenges. AI models need high-quality cleaned data to work well, but sometimes, the data is inconsistent, affecting performance. Due to this issue, it is important for industry leaders, regulators, and researchers to work together to create strong guidelines to ensure AI is used safely and effectively. [5] [6] In drug discovery, this area has its efforts but still lacks collaboration. Some programs have been developed between the NIH and Research Centers in Minority Institutions (RCMI), between 22 participating Universities, to help with the data-sharing gap. These include Science Collaborative for Health Disparities and Artificial Intelligence Bias Reduction (Schare), MIDRC, which is a multi-institutional data resource center hub for medical imaging data, and IMMPORT, a broader platform that includes more clinical trials, with some analysis for preclinical and cleaned assay data. [7] Another program more associated with Big Pharma is MELLODDY, where pharmaceutical companies share their models in a federated learning environment. [8] 

Pre-clinical Phase: AI has been shown to be utilized to reduce the amount of physical testing needed to analyze a candidate drug. However, I would like to briefly explain the difference between Computer-Aided Drug Design (CADD) and Artificial Intelligence for Drug Design (AIDD), as they tend to be confused at times in the industry. CADD typically includes similarity search, molecular docking, molecular dynamic (MD) simulations, quantitative structure-activity relationship (QSAR), etc. These all involve using computational tools and software to design drug candidates.  AIDD involves the use of artificial intelligence, machine learning algorithms, especially deep learning, and generative models, etc. [9] AIDD in preclinical aspects can be used to target identification, screen for small molecule hits and design of large proteins, lead optimization, predict the pharmacokinetics and pharmacodynamics (PK &PD), optimize drug formulations, predict reactions from drugs, and many more. [10] 

Here are some examples from the last five years that have shown progress in this area and a table to give perspective on other applications: In 2019, AI from Insilico Medicine was used to design a new molecule from scratch and validate its end-to-end activity in 46 days. [11] This is 15 times faster than in the past for pharma companies and has considerably reduced costs by up to 70%, saving $10-15 billion. [12] In 2020, Generative adversarial networks (GANs) were used to efficiently and quickly generate a broader spectrum of drug-like molecules and a diverse set of potential drug candidates. [13] Also, during 2020, Benevolent AI was used during the COVID-19 pandemic, and it was suggested that Baricitinib be repurposed to treat COVID-19. In seven months, the FDA granted it an Emergency Use Authorization (EUA). This drug showed a decrease in mortality for hospitalized patients when used together with remdesivir, and because of its effectiveness, it was quickly included in national and global guidelines for treating COVID-19. [14] In 2021, DeepMind’s AlphaFold algorithm was showcased in the 14th Critical Assessment of Protein Structure Prediction (CASP14) and received a Global Distance Test-Total Score (GDT_TS) of 92.4. This ranked AlphaFold2 first in the competition, demonstrating the dominance, accuracy, and reliability of the deep neural network (DNN) shown in Figure 1.

Figure 1: Schematic of the AlphaFold2 Deep Neural Network (DNN). The input sequence undergoes a genetic database search for multiple sequence alignment (MSA) and a structure database search for template pairing. The resulting MSA and pair representations are processed by the Evoformer (48 blocks) and the Structure Module (8 blocks) to generate a 3D protein structure. The process is iteratively refined through recycling steps to improve accuracy. [15]

This breakthrough in deep learning highlighted a 50-year challenge in protein folding and holds promise for the future of drug discovery to help with challenges in understanding the compositions of proteins and determining functions for future drug development. [15] In 2023, Natural Language Processing (NLP) programs such as IBM Watson, already known for analyzing vast amounts of biomedical data by extracting relevant information from scientific literature, have been enhanced to identify candidate genes and provide insight into potential pathomechanisms of traumatic heterotopic ossification. [16] Insilico Medicine also reported in 2023 that the FDA approved just 37 new drugs in 2022, compared to 50 in 2021 and 53 in 2020. To date, in 2024, over 100 drugs were approved. [17] 

Clinical Phases: Although most of AI's breakthroughs are in drug discovery and preclinical areas, clinical trials show some strides. For example, in 2020, the first AI-designed drug (DSP-1181) continued to Phase 1 clinical trials for the treatment of obsessive-compulsive disorder (OCD). [18] In 2022, Insilico Medicine's AI-discovered and AI-designed drug entered Phase 1 clinical trials in a record time of under 30 months. It is also important to note that it has progressed to Phase 2 in treating fibrosis. [19] Other examples on a more general scale where AI can help include finding the best patients for a trial by looking at medical records, genetic information, and other data to see who will most likely benefit from the treatment and help with the recruitment process [20] During the trials, AI can track the patient's health by continuously monitoring the data collected. It can detect and analyze any reported side effects and identify patterns and risk factors much faster than traditional methods, making the trial safer and more effective. [21] While there is much controversy about reducing cost and time, here is a short literature-reviewed table comparing my findings. 

FDA Review and Post-Market Phase: The process of getting a new drug approved by the FDA involves analyzing extensive data from multiple studies to show it is safe and effective. AI can speed up this process by quickly extracting key information from documents and analyzing past submissions to build a strong evidence package. After the approval, AI helps monitor drug safety and effectiveness in the real world. It can rapidly analyze data from the FDA’s adverse event reports and detect safety issues faster. AI also scans social media for discussions about side effects and reviews new research papers for emerging concerns. This real-time monitoring ensures ongoing drug safety and performance. [27] 
Conclusion: Traditionally, this process was laborious and costly, relying heavily on trial and error with an estimated cost of $ 2.8 billion.  AI has proven that it can make drug discovery and development faster, cheaper, and more efficient. It helps in many stages, from early research to monitoring drugs after they are on the market. The improvements seen so far suggest that AI will play a big role in the future of medicine. With the continued collaboration between scientists, companies, and regulators, AI should continue to advance and make a positive impact on the pharmaceutical industry and healthcare. However, keep in mind that there are still challenges that need to be faced, from obtaining high-quality data and improvements in the clinical areas to integrating more successful AI-generated drug candidates.

References