Selection of cost-effective yet chemically diverse pathways from the networks of computer-generated retrosynthetic plans

Abstract

A family of network algorithms allows the Chematica retrosynthetic platform to plan both cost-effective and chemically diverse syntheses.

Automatic retrosynthetic route planning using template-free models

“…Synthia (formerly Chematica), one of the most well-known, expertencoded, template-based retrosynthetic analysis tools, is a commercial program developed by Grzybowski and coworkers. 9,[13][14][15][16][17][18] This tool uses a manually collected knowledge database containing about 70 000 hand-encoded reaction transformation rules. 18 Based on human knowledge of organic synthesis and the encoding of organic rules over a period of more than 15 years, Synthia has been validated experimentally as an efficient toolkit for complex products recently.…”

Computational design of syntheses leading to compound libraries or isotopically labelled targets

“…Modern retrosynthetic planners (e.g., MIT's ASKCOS, 15 Waller's MCTS, 14 or our Chematica [17][18][19][20][21] ) differ in the origin of the underlying reaction rules (machine extracted vs. expert coded) and details of the search routines, but they all rely on the iterative expansion of parent/retron nodes into progeny/synthon nodes and on navigating (with the help of functions scoring individual reaction moves) the thus created bipartite graphs of synthetic options until simple and commercially available substrates are found. This procedure yields pathways that lead to a given target of interest and in which all chemical nodes are "synthesizable" (i.e., they are targets of at least one synthetic pathway tracing back to commercially available substrates) and the reaction nodes are "viable" (i.e., all their substrates are synthesizable).…”

“…denition of viability above and in ref. 21) of the root node will also have to make all substrates for the ultimate r dummy reactionin other words, the search algorithm will not stop until viable syntheses of all m i molecules in the library are identied. Of course, this dummy reaction is not supposed to skew the real searches for m i s in any way, and its execution cost is assigned as zero.…”

“…which most economical plans can be selected (by iteratively propagating the yield-scaled costs from substrates to products and thereby assigning a realistic, monetary cost to each plan). In addition to the selection procedures detailed in our recent publication, 21 a unique feature of library-wide design is that we wish to promote convergent synthetic plans that make use not only of common intermediates but also of the smallest number of different synthetic methodologiesto this end, a penalty is added to any new reaction type encountered in the solution graph. Chemically, such penalization ensures that it is more economical to perform the same reaction on a, say, 2 mol scale, rather than two different types of reactionsrequiring separate set-ups and likely different reagentseach on a scale of 1 mol.…”

“…Capitalizing on the advances in articial intelligence [1][2][3][4] and constantly increasing computing power, recent years have brought revived interest and signicant progress in the decades-old challenge of teaching computers the design of multistep organic syntheses. [5][6][7][8][9] Various platforms, differing in the underlying details of search algorithms and reaction-rule formats, have been developed [10][11][12][13][14][15][16][17][18][19][20][21] and one of these platforms, our own Chematica, [17][18][19][20][21] has been validated experimentally via successful execution of multiple routes leading to diverse, highvalue, medicinally relevant small molecules 18 and, more recently, natural products. 19 To date, a main effort in this emerging area of chemical research has been on algorithms designing syntheses of one specied target at a time.…”