Chemical Space

Based on: Chapter 1 of Foodinformatics, Applications of Chemical Information

to Food Chemistry, by Martinez-Mayorga K. & Medina-Franco, J.L.
Brandon Meza-González

September 11, 2018

Highlights

 Chemical Space (CS) is a set of molecules related between them by binary relations based on
some similarity measure.

 Similarity-Property Principle (SPP) connect similar molecules through bioactivities or pro-

perties. It can be applicable to any type of CS, and is used particularly in medicinal chemistry.

 There are three representations of CSs based in the position of molecules in space: (I) coordi-
nate based, (II) cell based and (III) graph based.

 Concepts like similarity cliffs have been introduced in order to describe relationships between
small changes in biological activity and large changes in similarity.

 Matched molecular pairs (MMPs) is a concept introduced due to differences generated by

representation or similarity coefficients.

 Comparison of Compound DBs is a task particularly suited by cell-based methods. Pearlman-

Smith (P–S) procedure address a deficiency of cell-based characteristics: “They are not specific
enough to afford a detailed comparative assessment, as they do not account for relationships
between the cells in collections being compared”.

 “Subset selection usually takes places in early screening while similarity searching or ligand
based virtual screening is typically used in subsequent follow-on screening activities.”

 The most common strategy used to generate initial screening sets is maximizing their diversity
by minimizing the similarity of compounds in set. Maximum dissimilarity can be sized by an
algorithm named Dfragall.

 Cell-based sampling schemes like simple, threshold based, proportional and property-based sam-
pling are used to obtain a subset of desired size and diversity.

 In order to enhancing the diversity and maintaining integrity of an existing collection, com-
pound acquisition is important to enhance assays and measurement of biological activity or
decompose over time.

 Ligand-protein docking, similarity searching based on 2D and 3D molecular descriptors are

approaches used to identify compounds with properties of interest.