How well does AI enable small molecule drug discovery?

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

Growth in the R&D pipeline of AI drug discovery companies

The authors of the article analyzed 24 companies that are mainly based on AI for drug discovery, of which the pipeline development trends of 20 companies from 2010 to 2021 can be reconstructed through information from public databases

▲ Changes in the pipeline growth of AI-driven drug discovery companies (a) and the world’s top 20 pharmaceutical companies (b) (Image source: Reference [1])

For comparison, the authors also assessed the internal R&D pipelines of the world's 20 largest pharmaceutical companies

However, how many of these preclinical projects can enter the clinical development stage, and how well they perform in clinical trials, still needs the test of time

Characteristics of the AI ​​Drug Discovery Pipeline

The authors conducted a further analysis of the current R&D pipelines of 24 AI drug discovery companies, and they found that only about a quarter of the projects in the R&D pipeline have specific target information

There may be several reasons for this tendency, including reducing development risk in internal pipelines, proving the viability of technology platforms, and addressing important challenges associated with well-proven targets (such as drug selectivity)

▲The target type (a) and the treatment disease field (b) of the pipeline of AI drug discovery companies and the world's 20 largest pharmaceutical companies (Image source: Reference [1])

However, some AI drug discovery companies have also discovered potential "first-in-class" compounds for innovative targets, such as protein phosphorylase SHP2, DNA helicase WRN and paracaspase MALT1, compounds discovered by AI.

In terms of therapeutic areas, most of the announced AI discovery projects are in the oncology and neurological fields, which may be related to the high unmet needs and the existence of multiple validated targets in these two fields

AI discovers chemical structure and characteristics of molecules

At present, the chemical structure data of the compounds discovered by AI is still very limited, so it is not yet possible to conduct systematic statistical analysis on them, but the analysis of some public data can also give us a glimpse of future trends

For example, the development of TYK2 inhibitors: TYK2 is a member of the JAK protein family, and a common problem with existing JAK inhibitors is limited selectivity, which affects their safety profile

Interestingly, when comparing AI-generated TYK2-specific inhibitors with less selective JAK inhibitors developed by traditional strategies, the authors found no significant difference in the chemical space occupied by the chemical structures of the two

The structures of two AI-discovered small molecules targeting the serotonin receptor have been published, and chemical space analysis shows that they occupy a similar chemical space to previously published drugs

▲Comparison of chemical spatial characteristics of compounds discovered by AI (green) and compounds discovered by traditional methods (Image source: Reference [1])

How quickly AI discovers molecules

One of the great goals of using AI to enable drug discovery is to accelerate the speed of drug discovery, for example, by rapidly discovering and validating targets, reducing the number of molecular design and optimization rounds and accelerating turnaround times

The author estimates the timeline of the research and development progress of some cooperation projects between pharmaceutical companies and AI drug discovery companies based on public data

▲Part of AI found the preclinical development speed of the therapy under study (the green dotted line is the industry average, picture source: Reference [1])

Epilogue

The authors say their analysis captures early signs of a rapidly approaching wave of AI empowerment, which has the potential to transform drug discovery

However, in other dimensions, such as reducing development costs, it is still too early to draw conclusions

References:

[1] Jayatunga et al.
, (2022).
AI in small-molecule drug discovery: a coming wave? Nature Reviews Drug Discovery, https://doi.
org/10.
1038/d41573-022-00025-1

(Original abridged)