Is Organic Chemistry Really Growing Exponentially?

Abstract

In terms of molecules and specific reaction examples, organic chemistry features an impressive, exponential growth. However, new reaction classes/types that fuel this growth are being discovered at a much slower and only linear (or even sublinear) rate. The proportion of newly discovered reaction types to all reactions being performed keeps decreasing, suggesting that synthetic chemistry becomes more reliant on reusing the well‐known methods. The newly discovered chemistries are more complex than decades ago and allow for the rapid construction of complex scaffolds in fewer numbers of steps. We study these and other trends in the function of time, reaction‐type popularity and complexity based on the algorithm that extracts generalized reaction class templates. These analyses are useful in the context of computer‐assisted synthesis, machine learning (to estimate the numbers of models with sufficient reaction statistics), and identifying erroneous entries in reaction databases.

AI-Driven Synthetic Route Design Incorporated with Retrosynthesis Knowledge

“… 8 , 13 , 14 However, knowledge-based approaches still require the great efforts of many experts, as the number of new reaction types discovered per year has been in the low few thousands. 15 …”

“…CASP approaches are generally classified into two types: knowledge-based , and data-driven approaches. , Knowledge-based approaches employ manually encoded (human-curated) transformations considering information, such as stereochemical and electronic effects . For instance, one excellent knowledge-based CASP application, Chematica (now rebranded as Synthia), provides a considerable discretion for chemists to perform retrosynthetic analysis based on their own ways of thinking using their own scoring functions (e.g., SMALLER, SELECTIVITY, and RINGS variables), and it is now used globally. ,, However, knowledge-based approaches still require the great efforts of many experts, as the number of new reaction types discovered per year has been in the low few thousands …”

“…8,13,14 However, knowledge-based approaches still require the great efforts of many experts, as the number of new reaction types discovered per year has been in the low few thousands. 15 In contrast, data-driven CASP aims to automatically extract knowledge related to transformations from numerous reaction records to discover synthetic routes. 16 Recent breakthroughs in deep learning (DL), 17,18 along with the availability of reaction records 19,20 and open-source codes, 21−24 have improved the core techniques of data-driven CASP such as 1-step (retro)synthetic reaction prediction 25−28 and multistep synthetic route searches.…”

Network search algorithms and scoring functions for advanced‐level computerized synthesis planning

“…By contrast, Chematica has consistently used its own set of network‐search algorithms with which it scores “synthetic positions” defined in terms of both the molecules it creates and reaction operations it performs. In doing so, the program pursues multiple search strategies supported by beam‐search‐inspired 44 priority queues as well as multistep strategizing routines 22,33 to seek more far‐sighted synthetic plans. Importantly, when MCTS and our algorithms were compared and used in conjunction with Chematica's other components, both algorithms performed well for simple targets (which is in line with a recent analysis of MCTS vs. an algorithm similar to Chematica's early versions [ 45 ]) but it was only the latter that was able to navigate routes to complex natural products—as opposed to MCTS which typically found no pathways 22 (likely because difficult syntheses cannot be finished by MCTS's rollouts).…”

“…23,24 By contrast, virtually all other teams pursued purely datadriven AI strategy, [25][26][27][28][29][30][31][32] largely motivated by the assumption that only automated extraction of rules can keep the pace of the "exponentially expanding" body of chemical knowledge (although, as recently shown, the growth is not nearly as rapid, and only a very small fraction of reaction classes have enough literature examples to allow for meaningful machine learning). 33 These efforts have been based either on (i) automatic extraction of rules from large reaction databases and then scoring the reaction "moves" by various types of neural networks, NNs, trained on these databases; or (ii) template-free methods using the so-called transformer NNs trained on large reaction repositories. 29,31 Given the recent feats of AI approaches in many other areas, the outcome of the "retrosynthesis rivalry" has been somewhat curious and not in AI's favor.…”