Is AI driving the wave of small molecule drug development?

Artificial intelligence (AI) has the potential to transform drug discovery

Some AI-native drug discovery companies have advanced drug molecules into clinical trials

AI has demonstrated several areas of value creation in drug discovery, including higher productivity (faster speed and/or lower cost), wider molecular diversity, and improved clinical success rates

Pipeline growth for AI drug discovery companies

The article focuses on analyzing 24 AI-native drug discovery companies with AI as the core of their drug discovery strategies

Figure 1 | The number of annual R&D projects shows the growth of AI-enabled drug discovery over time

In contrast, the combined internal development pipelines of the top 20 pharmaceutical companies contain approximately 330 disclosed discovery and preclinical projects, and approximately 430 projects in Phase I clinical development (using the same public data sources, excluding collaborative projects) ; Figure 1b)

The composition of the pipeline of an AI drug discovery company

Further analysis of the existing pipelines of 24 AI-native drug discovery companies in terms of therapeutic areas and target categories revealed that only about 1/4 of AI-enabled R&D projects can obtain detailed target information, but this part of the data The analysis of the set shows that AI-native drug discovery companies typically focus on identified target classes (Fig.

Figure 2 | AI drug discovery companies focus on identified target classes and therapeutic areas

This preference for proven targets may be driven by several factors, including AI companies' desire to de-risk their internal pipelines by focusing on targets of proven biology, demonstrate the viability of their technology platforms, and address important Challenges such as selectivity issues for well-characterized targets with abundant data, often including structural information

Despite these trends, there are several potential first-in-class AI-derived compounds for novel targets, including the protein tyrosine phosphatase SHP2, the DNA helicase WRN, and the paracaspase MALT1

In terms of therapeutic areas, most of the disclosed AI discovery projects are located in the oncology and central nervous system areas, likely due to the high unmet medical need and many well-characterized targets (Fig.

Chemical structures and characteristics of AI-derived molecules

Publicly available data on the chemical structures of AI-derived projects are currently limited

One example is TYK2 inhibitors

Interestingly, the authors observed no significant differences when comparing AI-derived TYK2-selective inhibitors with chemically less sterically selective JAK inhibitors developed by traditional strategies (Fig.

Figure 3 | Chemical space analysis of some AI-derived projects (Source: Nature Reviews Drug Discovery)

Data from some projects targeting serotonin receptors have also been disclosed

Taken together, these examples demonstrate that AI-enabled strategies can discover molecules comparable to traditional strategies' discovery efforts, with the potential to explore adjacent chemical spaces

The speed of development of AI-derived molecules

One of the greatest promises of AI-enabled drug discovery is an accelerated discovery timeline—for example, rapid target identification and validation, or fewer and faster cycles of molecular design and optimization
.

Measuring the discovery timeline using publicly available data is very difficult, but the authors found, based on the timing of patents, publications, and announcements, that multiple AI-enabled projects completed the entire discovery and preclinical process in less than 4 years ( Figure 4)
.
This data compares favorably to the industry's 5- to 6-year historical timeline, and given that AI is still in the exploratory phase, it may accelerate further as AI companies mature
.

Figure 4 | The preclinical development speed of some AI-enabled discovery projects (the dotted line is the industry average from target-to-hit to the start of clinical trials) (Source: Nature Reviews Drug Discovery)

Conclusion and Outlook

Drug discovery is a multi-dimensional, multi-step search and optimization problem
.
With its powerful new tools for solving complex problems, AI has the potential to play a major role in dramatically improving this process
.
Early indications are that a fast-approaching, AI-fueled wave has the potential to fundamentally change drug discovery
.

However, the impact of AI on different dimensions is different
.
First, there are early signs of efficiency and productivity gains
.
Most AI companies were established less than 10 years ago, but their preclinical production has already accounted for a considerable proportion of the TOP 20 pharmaceutical companies
.
The authors also found preliminary evidence of accelerated drug development
.
As for the rest, it's too early to draw conclusions
.
For example, it is currently difficult to assess its impact on costs, although the authors believe that systematically scaling AI in R&D could significantly improve costs
.
Most importantly, though, it remains to be seen whether the wave of AI discoveries will continue and translate into better drugs with clinical success
.

If successful, AI-enabled drug discovery will be a game-changer for drug discovery, especially small molecule drug discovery, potentially allowing it to "catch up" to other drug modalities that typically have faster development rates, such as monoclonal antibodies
.
This will impact how research and discovery institutions are organized and managed to realize the full potential of AI
.

References:

Madura KP Jayatunga et al.
AI in small-molecule drug discovery: a coming wave? Nature Reviews Drug Discovery.
2022