Chapter2: The History of ArtificialIntelligence and Chemistry

reading notes of《Artificial Intelligence in Drug Discovery》

---

• While many of the technological methods use experiments in so-called “wet” laboratories, the development and application of computational methods, often called in silico as opposed to in vitro or in vivo, have been in wide usage for many decades.

1.Artificial Intelligence in History & 2.The Winters of Artificial Intelligence

• This part is about the origin and history of AI in literature, science and theology.If you are interested in history, it is worth reading.

3.Chemistry Finding Artificial Intelligence

• The first recognised artificial intelligence application to chemistry was the Dendral (Dendritic Algorithm) system, begun in 1965. The primary aim of the Dendral system was to study the formation of hypotheses and scientific discoveries.

4.Synthesis Planning

• Although recent advances in applying artificial intelligence to challenges in synthetic organic chemistry, it is all too easy to forget the advances of those that came before us. A huge amount of significant work was conducted in the laboratories of both Corey and Ugi decades ago which led not only to advances in the chemical sciences, but also to advance algorithmic development in the computational sciences.

5.Predictive Modelling of Properties

• However, it is also important to realise that the synthesis and testing of a compound can also contribute to improving the prediction of properties, thereby having the side-effect of indirectly improving a project, while improving the predictive models in concert with the exploration and exploitation of the chemistry space under consideration.
• Alexander Crum Brown formalised the fundamental equation of a structure–activity relationship or model in his work of 1869.Crum Brown posited in this work that the physiological action of a compound is merely a function of its chemical constitution.
• It is interesting to note that QSAR models emerged around the same time as machine learning and artificial intelligence and the related fields have shared a number of advances over the past 60 years.
• While it would be impossible to consider all the different types of molecular descriptors that have been reported over the years—to that the reader is referred to the seminal work of Mauri, Consonni, and Todeschini16 that is an extensive compendium of many molecular descriptors that have been used in a variety of applications—it is perhaps useful to outline a few of the fundamental types of molecular descriptor that are used.
• Molecular descriptors can roughly be partitioned into knowledge-based and information-based descriptors. Knowledge-based descriptors are typically better for generating diagnostic models that enhance interpretability, while information-based descriptors are highly effective at generating accurate predictions, but are significantly more challenging to interpret.

6.Summary

• From Crum Brown’s definition of a predictive modelling in the mid-19th century onwards, together with formalisation of many aspects of mathematical graph theory, through early research in QSAR models and symbolic artificial intelligence in the 1960s, chemistry has been heavily involved in many advances of artificial intelligence.