Rewiring chemistry: algorithmic discovery and experimental validation of one-pot reactions in the network of organic chemistry.

In 2005 and 2006, we published the first reports on the representation of all synthetic knowledge as a giant network in which molecule “nodes” are connected by reaction “arrows” (Figure 1). In these early works, we focused on the topological structure and evolution of this network and demonstrated the scale-free network topology, existence of hub molecules central to organic synthesis, exponential growth of the network in time, correlations between molecular masses, trends in reactivity based on network connectivity, and more. While our analyses had little applicability to the everyday synthetic practice, we envisioned that such a junction between network theory and synthesis would one day be achieved. Now, we are reporting, in three consecutive communications, the extension of chemicalnetwork concepts into methods directly relevant to experimental chemistry: 1) discovery of one-pot reactions; 2) optimization of multiple reaction pathways, and 3) the detection and blocking of synthetic pathways leading to dangerous chemicals. The first communication in this series addresses one of the most important challenges in organic chemistry: namely, how to “wire” individual reactions into sequences that could be performed in one pot. One-pot reactions save resources and time by avoiding isolation, purification, characterization, and production of chemical waste after each synthetic step. Sometimes, such reactions are identified by chance or, more often, by careful inspection of individual steps that are to be wired together; even this latter process, however, is invariably subjective and depends on the knowledge and intuition of any individual chemist (or group of chemists) involved. Herein, we show that the discovery of one-pot reactions can be facilitated by computational methods. We first describe algorithms that identify possible onepot reactions within the network of all known synthetic knowledge and then demonstrate that the computationally predicted sequences can indeed be carried out experimentally in good overall yields. The experimental examples are chosen to “rewire” small networks of reactions around practically important chemicals: quinoline scaffolds, quinoline-based enzyme inhibitors, and thiophene derivatives. In this way, we replace individual synthetic connections with two-, three-, or even four-step one-pot sequences. The network of organic chemistry (NOC; Figure 1) is constructed from reactions reported in the chemical literature since 1779 and nowadays stored in chemical databases. Pruning the raw data to remove catalysts, solvents, substances that do not participate in reactions, and duplicate or incomplete reactions, leaves about 7 million reactions and about 7 million substances on which further analyses are based. This dataset is translated into a network by representing chemical substances as network nodes, and the reactions as arrows directed from the reaction s substrates to products. At first glance, this giant network of chemistry might look akin to the metabolic networks of biochemical reactions. In reality, however, metabolic networks are true chemical systems comprising reactions that can, in most cases, occur concurrently within the same reaction medium (that is, in Figure 1. a) A small (ca. 5500 nodes, ca. 0.1% of the total) fragment of the network of organic chemistry (NOC), where individual nodes represent the molecules and arrows represent reactions. The representation in b) has the reaction arrows colored by the times these reactions were first reported. This representation emphasizes the fact that NOC, by itself, is not a “coherent” giant chemical system but only a repository of reactions discovered separately, without regard for their mutual compatibility. At best, it can be said that there was a “coherent” interest in certain areas of chemistry (for example, synthetic activity around the Penicillin V node in the 1960s, following the first total synthesis).