 |
|
|
|
|
The need for the comparison of chemical reactions using computational tools is as basic as in the case of chemical structures. Techniques developed for the estimation of chemical (or other type of) similarity of molecules can be adapted to the similarity estimation of chemical reactions.
In order to define reaction similarity some basic concepts are introduced.
Simply speaking, chemical reactions transform one or more reactants to one or more products. Traditionally,
reactants are drawn on the left side of the reaction arrow, while products are placed to the right of it.
Thus reactants are often referred to as the left side of the reaction (and products as the
right side of the reaction).
One possible approach to characterize the transformation carried out (and thus to charactirize the reaction itself)
is to identify the changing atoms and the changing bonds in the reaction with respect to the
reactants and the product srtuctures.
An atom is changing if
Changing atoms and changings bonds define the reacting centre of the reaction. The reacting centre is specific to a particular type of reaction.
Atoms and bonds in the reacting centre are colored blue. (The numbers near reacting center atoms are the atom maps that associate the reactant atoms with the product atoms.)
Structural properties that are present in the reaction offer a natural approach to introduce the concept of
reaction similarity. Two
reactions can be considered similar if their product side and/or reactant side are similar. With this consideration,
reaction
similarity is reduced to molecular or structural similarity.
Nevertheless, another type of reaction similarity can be introduced by focusing on the reacting centre of the reaction. This
transformational similarity is less influenced by the particular reactant and product present in a reaction but
it is dominated by the reaction mechanism. Both of these types of reaction similarity are found to be useful in comparing
and matching reactions.
The degree of structural similarity is determined by the structural similarity of the reactants (products) present in the two reactions compared. The degree of transformational similarity can be examined at three distinct levels: strict, medium and coarse. Strict scale similarity compares the reactions with the broadest topological environment of the reacting centres. In contrast to this coarse similarity restricts similarity comparison to the bare reacting centres completely ignoring their neighborhoods. Topological distances introduced in the present implementation of transformational similarity are: 2, 1 and 0 according to the coarseness of similarity. These topological distances are interpreted as bond distances from atoms in the reacting centre.
Coarse similarity: changing atoms and bonds (i.e. reacting reacting center) are taken into account.
Medium scale similarity: atoms next to any atom in the reacting centers along with their bonds are taken into account. These atoms and bonds introduce further constraints and make the similarity comparison more strict.
The success of use of various types of fingerprints applied to the similarity calculations of chemical
structures encourages the
introduction of an analogous reaction fingerprint. It is apparent, that such fingerprint would support the
structural similarity assessment of reactions: the topological chemical fingerprint of the
reactants (products) can directly be compared. However, the transformational similarity should also be addressed,
nevertheless, at three different scales of coarseness.
Considering, that the transformational similarity was defined by topological properties of the reacting centres,
topological fingerprints appear to be viable choice for transformational similarity too. The reacting centre is a substructure
of the reactant (and the product), thus its topological fingerprint can be constructed and this fingerprint can be used to
represent the reacting centre. This concept can also be adapted to the
extended neighborhood of the reacting center, simply because the broader neighborhood of the reacting centre is
just another, larger substructure.
Based on the above considerations the structure of the reaction fingerprint is defined as follows:
The total length of the reaction fingerprint (in the present implementation) is 2048 bits. The above defined 8 segments of the reaction fingerprint are layed out in the schema below (segment sizes given in number of bit):
| 512 | 512 | 128 | 128 | 128 | 128 | 256 | 256 |
This reaction fingerprint enables both types of reaction similarity calculations, and with the expense of some extra storage space it makes the transformational similarity calculation efficient in all three predefined levels of coarseness.
Two types of reaction similarity calculations have been introduced: structural and transformational. Structural distinguishes the reactant and the product sides, while transformational relates to three levels of coarseness. With these considerations five metrics need to be introduced to efficiently estimate the five different cathegories of reation similarity. These metrics are as follows:
All of these metrics are based on the Tanimoto metric, consequently the degree
of similarity is expressed from 0 to 1.
ReactantTanimoto considers only the first quarter of the reactoin fingerprint that represents the reactants in the
reaction and ignores the rest of the reaction fingerprint. Therefore if estimates the structural similarity of the
reactants only.
ProductTanimoto takes the seconds quarter of the fingerprint that is associated with the products.
StrictReactionTanimoto takes the last two segments of the reaction fingerprint that represent the reacting centre of
both the reactant and the product side of the reaction with the broadest neighborhood and ignores the first 3/4 of the
reaction fingerprint.
Similarly, MediumReactionTanimoto applies the Tanimoto metric to the 5th and 6th segments; while
CoarseReactionTanimoto
takes the 3th and the 4th segments that encodes the reacting centre of the reactant and the product side, respectively.