top of page

A Snapshot into AI in Drug Discovery: Progress, Hurdles, and the Path Forward

Daniel Yuan
4 days ago5 min read

This image shows a scientist in a lab who is working on the developmental stages of organs-on-chips.


AI promised to revolutionize drug discovery, pledging faster development, reduced costs, and lowered risks. However, the reality has proven far more nuanced than initially expected. While AI is undoubtably making innovative strides in the industry, its true impact requires a closer look: where has AI succeeded, and where does it still fall short?  



The Current State of AI in Drug Discovery


AI models predict activity and generate novel insights by leveraging vast datasets to understand and replicate complex, heterogeneous biological systems. In drug discovery, AI’s predictive power depends on its ability to accurately model the intricate relationships between molecules, biological pathways, and disease mechanisms. AI-enabled companies often rely on a suite of interconnected models—ranging from machine learning to molecular simulations—that excel at specific functions to drive their predictive capabilities. While no single AI model or platform is currently capable of comprehensively replicating a biologic system, advances in data structuring, model training, and network integration continue to bring the field closer.


While no single AI model or platform is currently capable of comprehensively replicating a biologic system, advances in data structuring, model training, and network integration continue to bring the field closer.

Current Successes


AI models excel at analyzing immense volumes of information and finding patterns within existing data. As a result, these algorithms become increasingly adept at certain tasks and have already revolutionized aspects of drug discovery including: 


  • Enhancing and Repurposing Existing Therapeutics AI algorithms analyze extensive scientific databases to pinpoint connections, predict drug-disease interactions, and prioritize the most promising drug candidates while reducing both time and cost.  

  • Designing and Optimizing Synthesis Pathways Techniques like machine learning and retrosynthesis (the process of predicting chemical reaction pathways backward, starting from the target compound) accelerate the discovery and refinement of viable production routes for novel drugs, streamlining development. 

  • Virtual High-Throughput Screening (vHTS) Predictive models, trained on known drug interactions, rapidly simulate and evaluate molecular binding for a large number of candidates, significantly speeding up the identification of compounds with the highest therapeutic potential.  

  • Predicting Critical Drug Properties AI models leverage biochemical knowledge to predict critical drug properties such as toxicity and bioactivity (i.e., ADME), ensuring drug candidates align with safety and efficacy targets before entering clinical trials. 


When used correctly, AI plays a powerful role in drug discovery. Recently, an analysis reported that AI-discovered drugs have an 80-90% success rate in Phase 1 clinical trials, a sizeable improvement from the 76% industry average success rate for lead compounds. 


 AI-discovered drugs have an 80-90% success rate in Phase 1 clinical trials, a sizeable improvement from the 76% industry average success rate

Missed Marks and Growth Points


While the maturation of AI use in drug discovery has revealed areas of proven success, it has also unveiled missed marks and opportunities for growth: 


  • De Novo Drug Discovery: A commentary of the aforementioned analysis of clinical-stage AI drugs published in Science noted that none of the twenty-four drug candidates in clinical trials said to have been discovered by AI, targeted novel disease mechanisms, and several resemble already marketed therapies. These findings were substantiated by recent claims revealing that several drug candidates marketed as “AI-discovered” were optimized versions of drugs designed through traditional methods by other developers. These findings underscore the need for increased alignment between the genuine, methodical progress of AI, and its bold claims. 

  • Data Opacity: Companies in this space amass vast amounts of information—including, but not limited to, protein binding databases and best practices for generating and structuring diverse and relevant data—that are often kept hidden to maintain a competitive edge. While protecting intellectual property is essential, this secrecy limits opportunities to increase collaborative innovation that have the potential to unlock innovative therapies. 

  • Magnified Clinical Setbacks: AI drugs are pushed into the spotlight once they near or enter clinical testing. While clinical setbacks are an accepted reality of drug discovery, they are significantly magnified for AI-enabled companies who boast of their capabilities to slash costs and improve clinical trial outcomes. Consider Benevolent AI—once celebrated for its robust in-house drug development pipeline discovered by its cutting‐edge platform that reportedly identified a potential (now FDA-approved) COVID-19 treatment in just 90 minutes—recently slashed its entire development pipeline following several lackluster clinical performances.


While challenges around information sharing, AI’s actual impact on novel drug discovery remain, the progress of AI in drug discovery is unmistakable.  As such, recognizing both its incremental advancement and its limitations is essential to laying a solid foundation for AI’s future in drug discovery.  


A Look Ahead: Embracing All Results


While clinical failures for AI drugs are often seen as major setbacks, it is important to appreciate the value of negative outcomes.  


  • AI requires a continuous and iterative feedback loop to understand the quality of its output: in silico proposals must be validated in vitro and in vivo, positive and negative wet lab and clinical data must be properly analyzed and contextualized by humans, fed back into models, and tested.  

  • Training AI models on both successful and failed predictions helps improve accuracy and reliability and is a key foundation to unlocking AI’s transformative potential.  


Embracing all results in this back-and-forth cycle—computational modeling, lab work, animal and human studies—remains indispensable to AI’s development and progress.  


AI requires a continuous and iterative feedback loop to understand the quality of its output: in silico proposals must be validated in vitro and in vivo, positive and negative wet lab and clinical data must be properly analyzed and contextualized by humans, fed back into models, and tested.

The View from the Crow’s Nest


While the reality of AI in drug discovery is far more nuanced than initially expected, its transformational potential is undeniable.


Despite growing hesitation within AI drug discovery due to recent setbacks, incremental innovations are continuing to push AI’s capabilities to streamline traditional practices and unlock novel therapeutic potentials. By adopting structured approaches, emphasizing transparency, and integrating AI more seamlessly into existing research frameworks, the industry can pave a steady, credible path forward. 


 
If you are interested in learning more, get in touch at strategy@spinnakerLS.com. 

Spinnaker offers true partnership and comprehensive guidance to help leaders navigate the complexities of the Life Sciences industry and chart a path to success. From early-stage market assessment through commercial execution and ongoing lifecycle management, we deliver tailored solutions to ensure optimized practicable results.
 

Sources:  

 

 

bottom of page