[Yuriy/Adobe Stock]
The recent surge in adoption suggests that the technology is reaching the early majority stage of adoption. Bain noted in 203 that a substantial number of pharmas are expecting near-term cost efficiencies from the technology. A total of 40% of pharma execs are including expected savings from generative AI in their 2024 budgets, and 60% have set targets for cost savings or productivity boosts.
While AI seems increasingly ubiquitous in research labs, an AI-discovered candidate has yet to hit the market, highlighting the long road from innovation to approved therapy. This gap between current implementation and future potential mirrors the adoption curve of other transformative technologies, suggesting that AI in drug discovery is progressing along a well-trodden path from hype to practical reality.
Dave Latshaw
To unravel the current state and future potential of AI in drug discovery, we’ve reached out to Dave Latshaw II, Ph.D., former AI drug development lead at Johnson & Johnson and current CEO of BioPhy.
1. A growing number of AI-discovered drug candidates are in clinical trials
As AI techniques become de rigueur in pharma, a growing number of AI-discovered drugs and vaccines are advancing to clinical trials. “AI-native biotech companies and their pharmaceutical partners have entered 75 AI-discovered molecules into clinical trials since 2015, demonstrating a compound annual growth rate of over 60%,” Latshaw said. These molecules span a range of potential indications. Oncology is “particularly prominent,” Latshaw noted, representing about 50% of AI-discovered molecules in phase 1 and 2 trials.
2. AI-discovered molecules have better odds of success
Molecules identified through AI exhibit greater success in early clinical trials than those discovered using traditional techniques. “Phase 1 trials for AI-discovered drugs have shown success rates between 80-90%, significantly higher than the historical industry averages of 40-65%,” Latshaw said. “In Phase 2 trials, the success rate for AI-discovered molecules stands at 40%, aligning with historical averages.”
But while AI techniques are especially powerful at identifying drug-like properties and optimizing molecules for safety, further work remains in developing AI techniques to improve efficacy.
“Another consideration is the small sample size of molecules in the clinic which could create misleading statistics in these early days,” Latshaw said.
3. AI could potentially double R&D productivity
The promising early-phase success rates of AI-discovered molecules point to potentially substantial R&D efficiency gains. “The higher early-phase success rates of AI-discovered molecules suggest a potential doubling of overall R&D productivity,” Latshaw explained. “If these trends continue into phase 3 and beyond, the pharmaceutical industry could see an increase in the probability of a molecule successfully navigating all clinical phases from 5-10% to 9-18%.”
If this trend holds, it could have significant ramifications for the industry. “This productivity boost would enable companies to either reduce costs and resources for the same output or increase the number of new drugs brought to market within the same resource constraints,” Latshaw noted. However, he cautioned that realizing these benefits depends on whether “functions downstream of discovery can support the doubled workload.”
4. AI algorithms can help make sense of the omics data avalanche
Omics datasets, spanning areas such as genomics, proteomics, and metabolomics, are a valuable data source for drug discovery, but they can also be large with individual studies reaching terabytes to petabytes in size. Thankfully, machine learning techniques can help find order in the chaos. “Knowledge graphs, for instance, integrate and analyze diverse biological data to reveal complex relationships between genes, proteins, and diseases, facilitating the identification of novel drug targets and biomarkers,” Latshaw said.
From a drug discovery perspective, AI algorithms can help identify drug targets and biomarkers more quickly than traditional methods. They can also shed light on disease mechanisms while providing data to train models that can better predict drug efficacy, toxicity and patient responses.
5. New innovative tools emerging
In recent years, interest in generative AI models has exploded. In drug discovery, the pharma industry has explored such models’ potential to assist with designing small molecules. “These models can generate novel chemical structures with desired properties, optimizing for factors like binding affinity, selectivity, and pharmacokinetic profiles,” Latshaw said. “By simulating molecular interactions in silico, these tools significantly reduce the time and cost associated with traditional drug discovery methods.”
Predictive analytics can help with compound prioritization, helping researchers decide which compounds to pursue. “AI models analyze historical data on drug efficacy, toxicity, and clinical trial outcomes to predict the success likelihood of new compounds,” Latshaw said. “These insights help researchers focus their efforts on the most promising candidates, thereby increasing the efficiency of the drug development pipeline.”
The impact extends beyond the lab into trials. “By predicting patient responses and identifying optimal patient populations, these tools enhance trial efficiency and success rates,” Latshaw noted. In addition, AI algorithms can continuously analyze trial data to recommend real-time adjustments to optimize for safety and efficacy.
In an adjacent space, AI techniques are also employed to repurpose existing drugs for new therapeutic uses. “By analyzing existing drug data, AI models can identify new indications for approved drugs, speeding up the development process and reducing costs associated with bringing new drugs to market,” Latshaw said.
6. Big Pharmas increasingly forging strategic partnerships to tap AI expertise
When adopting novel technologies, a fundamental calculus is to determine to which tools to build and which existing products and services to buy. In the pharmaceutical landscape, many companies are opting to lean on strategic partnerships while also building up their rosters of AI-savvy employees and enhancing their data proficiency. For instance, Eli Lilly and Novartis have entered into significant partnerships with Isomorphic Labs, the Alphabet subsidiary, for AI-driven drug discovery. NVIDIA is also increasingly focused on the drug discovery segment.
“We’re seeing a significant shift in how Big Pharma approaches AI integration,” Latshaw observed. “Companies are increasingly opting for strategic partnerships over acquisitions, allowing them to tap into specialized AI expertise without the complexities of mergers and acquisitions.”
Latshaw notes that these partnerships raise important considerations beyond their immediate benefits. “Although not addressed explicitly, another item for consideration is data privacy issues stemming from global concerns about OpenAI’s approach to model training and the use of data that has not been approved for training,” he said.
“With the amount of new public data for model training diminishing rapidly, this highlights new approaches that companies like OpenAI might use to gain access to private data to continue improving their model sophistication and competitive moat,” he concluded.
Filed Under: clinical trials, Drug Discovery, machine learning and AI
