Split-and-combine method

Scheme 79. The split and combine method of preparing a combinatorial small molecule library...

A further improvement of this approach, to create increased molecular diversity, was applied to the synthesis of a soluble peptidomimetic combinatorial library of about 60 000 compounds [22]. This library generated using the split and combine method is based on a dipeptide scaffold, which incorporates 50 different L-, D- and unnatural amino acids,... 

To demonstrate the potential of this method, a model library of isoxazolidines on Rink amide resin was synthesized by the split-and-combine method [25] (Scheme 16.1). The polymer-bound library obtained consists of a total of 18 compounds. If one considers the different building blocks of the synthesis, a nitro group or the carboxamidomethyl residue is found in 50 % each of the compounds, a cyano, N,N-dimethylamino, or trifluoromethyl group in 33 % each and a succinimide, N-methylsuccinimide, or sulfone group in 33 % each. 

Scheme 16.1 Solid-phase synthesis of isoxazolidines according to the split-and-combine method, (a) Distribution of the resin into two equal portions, coupling of bromocarboxylic acids with N,N -diisopropyl-carbodiimide (D1C), combination of the resin, substitution with hydroxylamine. (b) Distribution of the resin into three equal portions, condensation with three different aromatic aldehydes to the corresponding nitrones, combination of the resin, (c) Distribution of the resin into three equal portions, cyclo-addition with three different dipolarophiles to isoxazolidines, combination of the resin.

When applied together with a LC-MS analysis, IR mapping allows a statistical investigation and thus a rapid quality control of the library synthesized by the split-and-combine method. The decisive advantage of this technique is that 1R analysis, in contrast to mass spectrometry, can be carried out without destruction of the sample, and with spatial resolution. 

In related work a library of 1,458 peptide ligands and various metal salts was tested in hydrolysis reactions of (p-nitrophenyl)phosphates.35 An active substructure composed of polymer-bound histidine in combination with Eu3+ was identified by further dissecting the original hit structure. It needs to be pointed out that catalytically active polymer beads can also be tested for catalytic activity using IR-thermography. In a seminal paper this was demonstrated using 7,000 encoded polymer beads prepared by split-and-pool methods, specifically in the metal-free acylation of alcohols.36... 

As with other applications of combinatorial chemistry, libraries of catalyst candidates are mainly generated either by parallel synthesis (Scheme 3, middle) (6) or by the split-and-combine approach (Scheme 4) (7). Parallel (in many instances automated) synthesis has the advantage that the identity of the synthesis products (e.g., ligands) is known, and that existing methods of solution-phase synthesis (and analysis) can be applied with only little modification. 

In the split-pool procedure, also known as the portioning-mixing method, the solid support is first divided into as many equal portions as the number of amino acids in the peptide s sequence [74,75,83], Each portion is coupled individually to only one amino acid. All portions of the resin are mixed, and the entire process of splitting and combining is repeated until all amino acids have been combined. An exorbitantly large number of derivatives can be synthesized at a time. 

Figure 2.4. Principle of the split-and-pool method for preparation of large libraries of mixed-metal oxide catalyst beads (black dots) are covered in liquid phase by the components A, B, and C, then they are separated from the liquid and united before being exposed again to A, B, and C. After only three steps, 27 combinations of components have been prepared.

Although this method has not found as much wide acceptance when referenced to use by designers or controversial discussion in the literature, it nevertheless allows a direct approximate solution of the average multicomponent system with accuracy of 1-8% average. If the key components are less than 10% of the feed, the accuracy is probably considerably less than indicated. If a split key system is considered, Scheibel reports fair accuracy when the split components going overhead are estimated and combined with the light key, the badance considered with the heavy key in the L/D relation. 

Additionally, Breiman et al. [23] developed a methodology known as classification and regression trees (CART), in which the data set is split repeatedly and a binary tree is grown. The way the tree is built, leads to the selection of boundaries parallel to certain variable axes. With highly correlated data, this is not necessarily the best solution and non-linear methods or methods based on latent variables have been proposed to perform the splitting. A combination between PLS (as a feature reduction method — see Sections 33.2.8 and 33.3) and CART was described by... 

Table I lists the comparative parameters for the various indochinite spectra. Two methods were used in preparing these samples. The first two samples listed were prepared by grinding the indochinite specimen and binding the powder with water glass. The other samples were sliced with a diamond saw. The two spectral lines are given with their position, width, height, and area. The quadrupole splitting and isomer shift are listed in the columns labeled QS and IS. (The isomer shift is really a combination of isomer shift and temperature-dependent shift, and the values are relative to iron in palladium.) The raw data points were fitted with a two-peak Lorentzian using an IBM 7094 least-squares fit.

The nonlinear resilience analysis methods of the previous few sections, although rigorous, are limited to rather specific situations (Saboo et al., 1987a,b) minimum unit HENs with piecewise constant heat capacities (but no stream splits or flow rate uncertainties), minimum unit HENs with stream splits (but constant heat capacities and no flow rate uncertainties), or minimum unit HENs with flow rate and temperature uncertainties (but constant heat capacities and no stream splits). Although it might be possible to combine these resilience analysis methods, the combined method would still be limited to HENs with a minimum number of units, and it would only be a sufficient test for resilience (at least for HENs with stream splits). 

All the spectroscopic approaches applied for structural characterization of mixtures derive from methods originally developed for screening libraries for their biological activities. They include diffusion-ordered spectroscopy [15-18], relaxation-edited spectroscopy [19], isotope-filtered affinity NMR [20] and SAR-by-NMR [21]. These applications will be discussed in the last part of this chapter. As usually most of the components show very similar molecular weight, their spectroscopic parameters, such as relaxation rates or selfdiffusion coefficients, are not very different and application of these methodologies for chemical characterization is not straightforward. An exception is diffusion-edited spectroscopy, which can be a feasible way to analyze the structure of compounds within a mixture without the need of prior separation. This was the case for the analysis of a mixture of five esters (propyl acetate, butyl acetate, ethyl butyrate, isopropyl butyrate and butyl levulinate) [18]. By the combined use of diffusion-edited NMR and 2-D NMR methods such as Total Correlation Spectroscopy (TOCSY), it was possible to elucidate the structure of the components of this mixture. This strategy was called diffusion encoded spectroscopy DECODES. Another example of combination between diffusion-edited spectroscopy and traditional 2-D NMR experiment is the DOSY-NOESY experiment [22]. The use of these experiments have proven to be useful in the identification of compounds from small split and mix synthetic pools. 

There are several ways to reduce the amount of experimental work, and combining them can be very effective. First try to understand how the product is to work. We have already described that for our example. This helps to split off parts of the work that are independent - parts that can be studied on their own. Another way to reduce the number of experiments is to use models of the process or parts of it to find suitable starting values. Then we have seen that systematic coverage of the design space is not an efficient way of finding the best solution there are better search methods. Finally, it is often possible to do many experiments simultaneously. Together these techniques can reduce the time needed enormously, and that is just as well. 

---