Distributed Heuristic Synthesis Search

Abstract

The Stony Brook SYNCHEM program, a well-established intelligent problem-solving system for synthesis route discovery, recently has been upgraded to execute in a network of multiprocessor workstations under both Linda tuple space and PVM message passing protocols. We describe the implementation and report experimental results. In addition, we discuss the effect of parallelizing SYNCHEM's search algorithm on the global best-first heuristic exploration of the synthesis problem space. Though reasonably wide-ranging and inclusive, sequential exploration exhibits a pronounced depth-first flavor. Parallelization introduces a countervailing breadth-first tendency that alters the region of the problem space explored by the search engine with mixed consequences, the positive clearly outweighing the negative.

Computer‐Assisted Synthetic Planning: The End of the Beginning

“…SYNLMA [7] was an effort by P. Y. Johnsons group from the Illinois Institute of Te chnology and was significant because it separated the knowledge-base from its "reasoning component" based on logical operations to be applied during retrosynthesis.U nfortunately,t he program ran into the "combinatorial explosion" problem generating excessively large retrosynthetic trees which it could not meaningfully prune.Itd isappeared from the scene already in 1989. SYNCHEM [8] and its successors were under development at Stanford/Stony Brook already at the time of LHASAs initial publication, but the program came to light only in 1977. Thet ruly innovative aspect of this approach-especially at the times when modern computing was in its infancy-was that it attempted to construct and explore (with BFS-like Angewandte Chemie Reviews 5918 www.angewandte.org searches) full retrosynthetic trees leading to few-thousand memory-stored commercial products and using on the order of few hundred expert-coded (but general-level) transforms.…”

“…The computer also identifies more synthetic plans for this compound. All analysesw ere performed in Syntaurus with CSF = SMILES_LEN 3/2 + SMILES_LEN;R SF = 40 + 60·PROTECT + 50·CONFLICT 2 and tracing the searches to commercially available reagents with MW < 125. some relevant reaction rules/transforms were missing in the machines knowledge base (e.g., as in LHASA [5] or SYN-CHEM [8] ). In contrast, modern programs such as ARChem, [41] IC SYNTH , [42] or Syntaurus rely on complete or nearly complete reaction databases-where they go wrong is in predicting reactions which on paper might look acceptable but will not work in the laboratory.Aswenarrated earlier,the majority of these problems can be avoided at the level of properly coded reaction rules taking into account admissible substituents, stereochemistry,r egiochemistry,p rotection group chemistry and reactivity conflicts,aswell as electronic and most of steric effects,t he latter at the scale of the reaction motif being coded.…”

“…Although only few of the early synthetic-planning programs are still available today,their importance for the development of the field should not be underestimated, as they galvanized the creation of various types of machine-readable molecular notations (e.g., SMILES) [4] and also structure editors (e.g., Stewart Rubensteins ChemDraw emerging from Coreys LHASA effort [5,6] ). Programs such as P. Y. Johnsons SYNLMA [7] contributed by introducing the elements of formal logic into synthetic planning,G elernters SYN-CHEM [8] was ap ioneering effort of searching expanding trees of synthetic possibilities,a nd Hanessians CHIRON [9] emphasized the recognition of similar structural patterns in the targets and in the substrates.T hese are all essential aspects of synthetic planning-the question remains,t hough, Sara A. Szymkuć graduated summa cum laude from the Faculty of Chemistry of the Warsaw University of Technology with thesis on multicomponent reactions applied to peptide mimics. She is currently agraduate student at the Institute of Organic Chemistry of the Polish Academy of Sciences in Warsaw.Her scientific interests focus on computer assisted organic chemistry,chemical networks, and multicomponent reactions.…”

Computergestützte Syntheseplanung: Das Ende vom Anfang

“…Wie bei allen früheren Programmen war die Stereochemie ein Hauptproblem, und die Regiochemie wurde nicht betrachtet. Die letzte Publikation zu SYNCHEM erschien 1998 und beschrieb Arbeiten zur Parallelisierung des Codes. Danach schien sich SYNCHEM anderen Retrosyntheseprogrammen im Walhall der computergestützten Chemie angeschlossen zu haben.…”

RetroTransformDB: A Dataset of Generic Transforms for Retrosynthetic Analysis

“…It should be noted that the quality of results from rule-based system strongly depends on the available reactions for the purpose of the retrosynthetic analysis-the so-called transforms [4]. And while many retrosynthesis software systems are based on manually coded rules [5][6][7][8], some systems [4,9] attempt to automate the rule (transforms) generation process [10] in order to cover more reactions. Applying such an approach is certainly attractive, but the depth of the predictive models that use it strongly depend on the reaction databases they are working with [11].…”