In recent years, the exploration and exploitation of chemical space have primarily relied on traditional approaches rooted in chemical intuition or on large databases. While these methods have provided significant insights, they often exhibit a high level of bias, limiting the generalizability of the findings. This bias arises from the inherent constraints of human intuition and the limited scope of existing chemical databases, since variants of synthesized compounds are often highly self-similar. This restriction in chemical space likely a) makes a search harder, since useful compounds need to be identified within a smaller subset of the available compound space and b) it may overlook novel and interesting concepts. Somewhat worse, human intuition is hardly updated if never challenged with more diverse chemistry, which may yield self-fulfilling design approaches. Consequently, there is a pressing need to develop strategies that enable a more diverse and comprehensive exploration of chemical space.

Recent advancements in computational methods offer a promising avenue to address this challenge. Unlike experimental approaches, computational techniques can efficiently explore vast chemical spaces by simulating and modeling new systems without the constraints of physical experiments. This ability makes computational methods an ideal starting point for developing diversification strategies in chemical space exploration. Many databases push for more diverse datasets for molecules [1,5], reactions [6-8] and materials [9], but coverage of element diversity is low even in these cases [2]. Machine learning representations have been tailored towards robustness for diverse systems especially in generative modelling applications [3] but foundation models for (materials) chemistry [4] still operate on biased and sparse data.

The core question on the workshop will be: How can we find fast and reliable strategies for a more diverse exploration of possibly relevant compounds and materials? What are necessary requirements for those strategies to be impactful and practical - both from design and application perspectives?