Let’s dive into the fascinating realm of pharmaceutical innovation, where discovering new drugs unfolds as a challenging, time-intensive, and expensive journey. It typically spans 12-15 years, involves a hefty 2.5 billion USD investment per drug candidate, and, unfortunately, sees around 90% of them failing in late-stage clinical trials. 

The industry’s paramount challenge is now to accelerate this process and foresee the success or failure as early as possible in the preclinical stage to make the process more efficient. In this pursuit, artificial intelligence (AI) emerges as a game-changing tool, already displaying significant promise in expediting and enhancing multiple phases of drug discovery.

Before delving into the technical intricacies of AI’s impact, let’s explore recent financial dealings between major pharmaceutical players and AI-driven startups. Take, for example, ISM3091, a small-molecule inhibitor developed by the AI-driven startup Insilico Medicine, licensed by Exelixis for 80 million USD in August 2023. Merck, not to be left behind, sealed a deal with BenevolentAI and Exscientia for an upfront 30 million USD and milestone payments reaching 1 billion USD. This trend extends to big pharma giants investing heavily in integrating AI tools into their internal discovery pipeline. Johnson & Johnson, for instance, has poured hundreds of millions into its AI-driven drug discovery endeavors, hiring 6000 expert data scientists and establishing a cutting-edge research facility in the San Francisco Bay area, inaugurated in September 2022. Despite over 70 drug modalities entering human clinical trials discovered using AI-driven tools, the question remains: Will they pass the clinical trial challenge, or will more than 90% fail just like the traditionally discovered ones? The answer awaits us in the years to come.

While still in its infancy, AI-driven drug discovery champions the idea of significantly reducing the time to take a drug from preclinical discovery to human clinical trials. However, a reality check is in order. Consider Insilico Medicine, a leading AI biotech based in Hong Kong, the US, and Canada. They identified a novel target for treating idiopathic pulmonary fibrosis (IPF) using their AI-based Biology42 platform, designed a novel small molecule inhibitor through their Chemistry42 platform, and entered Phase I human clinical trials within 30 months. 

This end-to-end AI-generated molecule showcased positive results in Phase I, earned Orphan Drug Designation from the US FDA, and is currently undergoing evaluation in Phase II human clinical trials. It is a testament to AI’s potential in trimming the preclinical discovery time from 6 to 2.5 years. While clinical testing maintains its traditional pace, the surge in candidates entering trials may herald a new era. A recent Boston Consulting Group (BCG) report estimates AI’s potential to save 25-50% of the time and expense in preclinical drug development, signaling a transformative impact on the pharma industry. However, numerous hurdles lie ahead as AI-discovered drugs progress further in the clinical development pipeline.

The drug discovery odyssey typically commences with target identification – a biological entity like a protein, enzyme, DNA/RNA, or gene with a causal link to the disease mechanism. Take chronic hepatitis B as an example, where patients exhibit elevated levels of hepatitis B virus surface antigen (HBsAg) in the bloodstream. To treat this, the key is to reduce the HBsAg levels secreted from infected hepatocytes. The challenge is determining which viral/host protein/gene to target to reduce serum HBsAg levels. Here’s where AI, fueled by extensive experimental datasets, comes to the rescue, establishing networks between biological entities and uncovering novel targets that might otherwise be missed due to the dataset’s complexity and size. Once identified, the target undergoes validation in cell-based experiments and animal models. 

Another critical aspect is assessing its druggability – can it be targeted with a small molecule drug or a biologic like an antibody? AI-based tools predicting druggability based on target structure are currently under development. The advent of AlphaFold even enables predicting the structure of a protein from its sequence, even if no experimental structure exists. This prediction aids in determining the druggability of the protein.

Then, the task shifts to finding a needle in the haystack – a small molecule from the multi-billion size chemical space that can bind to the target protein. AI-based tools accelerate the virtual screening of vast molecular libraries, making the process faster and more accurate. 

Generative AI takes it further by producing novel molecules with desired medicinal chemistry attributes, bypassing the need to sift through the haystack. Several AI-based drug discovery startups, including BenevolentAI, Exscientia, Insilico Medicine, Nimbus Therapeutics, Pharos iBio, Recursion Pharmaceuticals, Relay Therapeutics, Schrödinger, Structure Therapeutics, and Valo Health, have successfully employed generative algorithms to derive promising molecules now undergoing clinical trials. Merck’s introduction of AIDDISON, an AI-driven drug discovery tool, adds momentum. It can generate new molecules and predict their synthesis route through their existing AI-driven retrosynthesis software, SYNTHIA. Insilico Medicine’s Medicine42 tool has also evolved, now predicting clinical trial outcomes, designing more effective clinical trials, and potentially saving millions on flawed designs, leading to inconclusive clinical data.

AI-driven drug discovery initiatives aim to streamline the Design, Make, and Test cycle for drug-like molecules – making it more efficient, faster, accurate, and automated. The vision encompasses precisely predicting novel drug targets based on experimental biological data, generating new molecules, predicting their medicinal properties, and determining the most efficient manufacturing processes. Automated and integrated robotic systems will further take care of chemical synthesis, purification, and characterization, similar to what Chemify, a startup founded by Prof. Lee Cronin at the University of Glasgow, has already started doing. 

A decade ago, all of these might have seemed like science fiction, but today, with tangible technologies overcoming millions of limitations, breakthroughs appear on the horizon. The hope is to witness a revolution in the pharmaceutical industry, transforming it into a more efficient and productive drug discovery engine.

Imrul Shahriar recently obtained his Ph.D. in Chemistry from the Department of Chemistry at Purdue University, West Lafayette, Indiana, USA. He worked on the development of antiviral immunotherapies with a primary focus on creating effective therapeutic strategies against viral infections caused by influenza and other enveloped viruses. He is currently working as a Postdoctoral Research Associate at Purdue University and serving as a Research Consultant at Eradivir Inc.
imrul.shahriar@gmail.com