Activity cliffs in drug discovery: Dr Jekyll or Mr Hyde?

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Machine Learning

Background:

• Activity cliffs, regions of rapid change in biological activity with small molecular structure changes, are well-documented in medicinal chemistry.
• Their positive impact on drug discovery is recognized, aiding chemists in exploring chemical space.
• However, the negative influence of activity cliffs on the development of predictive machine learning models and similarity-based approaches is less understood.

Purpose of the Study:

• To review the dual role of activity cliffs in both medicinal chemistry and computational drug discovery.
• To highlight the understudied detrimental effects of activity cliffs on machine learning algorithms.
• To discuss potential solutions for mitigating the negative impact of activity cliffs in computational methods.

Main Methods:

• Literature review focusing on activity cliffs in drug discovery.
• Analysis of the impact of activity cliffs on quantitative structure-activity relationship (QSAR) and machine learning models.
• Exploration of similarity-based approaches in the context of activity cliffs.

Main Results:

• Activity cliffs offer opportunities for medicinal chemists to navigate chemical space effectively.
• Activity cliffs pose significant challenges for building robust predictive models in computational chemistry.
• Existing research has not sufficiently addressed the negative consequences of activity cliffs on machine learning.

Conclusions:

• Activity cliffs have a complex, dual impact on drug discovery, benefiting medicinal chemistry while complicating computational approaches.
• Further research is needed to develop effective strategies for handling activity cliffs in machine learning and similarity-based drug discovery.
• Addressing the 'ugly face' of activity cliffs is crucial for advancing predictive modeling in drug discovery.