Cheminformatics: storing and searching chemicals

Cheminformatics uses computers and information science to solve problems in chemistry, combining chemistry, physics, biology, mathematics and statistics. It is central to computer-aided drug design (CADD).

Cheminformatics specialists work with the physical properties of compounds, their 3-D and crystal structures, the pathways of chemical reactions, and huge virtual libraries of compounds, many of which have never been made yet. Virtual screening searches these libraries by computer to pick the most promising candidates for real laboratory testing.

The largest collection of chemical information is CAS (Chemical Abstracts Service), part of the American Chemical Society: more than 219 million substances and more than 70 million sequences. Other databases include PubChem, ZINC, ChEMBL, NCI, ChemSpider, BindingDB, DrugBank, PharmaGKB and SuperDrug.

A picture of a molecule is of little use to a computer for searching. So structures are stored as molecular graphs: each atom is a node (a point) and each bond is an edge (a line joining two points). The same idea can map whole metabolic pathways, such as glycolysis.

Chemicals can be searched in three ways:

• By property: for example, all compounds with a boiling point in a given range.
• By substructure: finding all molecules that contain a particular group (such as a functional group), by matching part of a graph inside larger graphs (subgraph matching). To speed this up, molecules are first filtered using bitstrings (codes of 0s and 1s).
• By reaction: finding reactions, by matching which atoms in the reactants become which atoms in the products (atom mapping).