Synthesis search strategy

Synthesis is the step in design where one conjectures the building blocks and their interconnection to create a structure which can meet stated design requirements. This review paper first defines chemical process synthesis and indicates the nature of the research problems—to find representations, evaluation functions and search strategies for a potentially nearly infinite problem. It then discusses synthesis research and the most significant results in each of six areas—heat exchanger networks, separation systems, separation systems with heat integration, reaction paths, total flowsheets and control systems. 

The search strategy utilized both keywords and section searches to produce approximately 8700 hits between 1982 and 1993. Refinement reduced this number to around 2000 references and manual scanning further reduced the number to about 1200. Very few reports of synthesis results are included except where these reports clarify the topic. Review articles that focus on specific areas are covered in this chapter and should be consulted for more detail. 

The distinction between the Selection of transforms by Applicability and by Relevance is an important one when considering the strategies one might employ to search for synthesis sequences. The search in a Planning Space has the characteristic that the search "leaps" into some intermediate point in the synthesis sequence and establishes an "island" and the solution search could then proceed from the island to the target molecule in the forward direction or from the island backward in the retrosynthetic direction toward available molecules. The significance of this ability to leap has been explored in other task areas than synthesis search and has been found to be a powerful tool in converging on solutions rapidly. Its utility for synthesis search remains to be shown and for now can be illustrated only in terms of examples. 

Search Strategies for the task of Organic Chemical Synthesis. Advanced Papers of the Third International Joint Conference on Artificial Intelligence, Stanford, 1973. 

Developing a suitable synthesis strategy for a target compound by searching for synthesis precursors, starting materials and synthesis reactions... 

We can obtain a crude estimate the time required for a precise quantum mechanical calculation to analyse possible syntheses of bryosta-tin. First, the calculation of the energy of a molecule of this size will take hours. Many such calculations will be required to minimise the energy of a structure. A reasonable estimate may be that a thousand energy calculations would be required. Conformation searching will require many such minimisations, perhaps ten thousand. The reactivity of each intermediate will require a harder calculation, perhaps a hundred times harder. Each step will have many possible combinations of reagents, temperatures, times, and so on. This may introduce another factor of a thousand. The number of possible strategies was estimated before as about a million, million, million. In order to reduce the analysis of the synthesis to something which could be done in a coffee break then computers would be required which are 10 times as powerful as those available now. This is before the effects of solvents are introduced into the calculation. 

Such techniques mean that the chemical literature may be used more effectively, and that its use can be partially automated. Might this lead to a way of automating organic synthesis To make most molecules there are many strategies which may be successful. If each reaction of each strategy can be evaluated for similarity to a reaction recorded in the literature, it should be possible to develop a route to most molecules by mechanically searching the chemical literature, so that suitable precedent is found for every transformation. 

For both these reasons, a strategy based simply on literature searching is unlikely to be competitive with the best synthetic chemists, who would, of course, use the literature to aid their synthetic designs. It may seem, then, that organic synthesis will remain a skill in which computers cannot compete with humans for some considerable time to come. However, this is not necessarily so. 

The ionizability of compounds affects other parameters such as solubility, permeability, and ultimately oral bioavailability, so it may be important to track changes in the pka of new compounds. Calculated pka values can be used when planning the synthesis of new compounds, but it is also a good idea to confirm these values experimentally. An example where this strategy can be useful is in the search for bioisosteric replacements for a carboxylic acid group. 

Solid-phase synthesis of biomolecules, of which peptides are the prime example, is well established. The search for more effective therapeutic agents creates a need for different strategies to synthesize peptides with C-terminal end groups other than the usual carboxylic acid and carboxamide functionalities. Methods described herein are readily generalized to small nitrogen-containing organic molecules. 

Rodgers, David and W. T. Wipke. Artificial Intelligence in Organic Synthesis. SST Starting Material Selection Strategies. An Application of Superstructure Search, J. Chem. Inf. Comput. Sci., 24, (1984), pp. 71-81. 

---