Hydrogen bears the distinction of being both the most common element in the universe and one of the most predicable elements on the periodic table. Outnumbering carbon in many organic molecules, hydrogen rarely participates in anything more exciting than hydrogen bonding or acid-base reactions. For the most part, the monovalent hydrogen atoms studding the average organic molecule are ignored.

A cheminformatics toolkit, like most software, is made of layers. The bottom layer deals with graphs and their manipulation. Above that sits a tightly coupled layer dedicated to the representation of molecular objects. All of a toolkit’s strengths and limitations ultimately evolve from these two layers. Previous articles have described minimal APIs for Graph and Molecule.

SMILES notation supports the reading and writing of tetrahedral stereochemical configuration. Usually the job of dealing with this notation falls to software. But sometimes it falls to you, the chemist or software developer. This article explains the SMILES stereochemical notation system in detail.

Over the last several decades, cheminformatics has managed to produce a few dozen different ways to exchange molecular encodings and metadata. Despite ongoing experimentation with high-level encoding techniques, these formats are mostly devoid of new ideas around bonding. To the extent that bonding relationships are encoded at all, they occur almost universally as the union of exactly two atoms.

Three line notations find regular use in cheminformatics: SMILES; InChI; and IUPAC nomenclature. Of these three, only SMILES finds widespread use in machine-to-machine chemical structure exchange. That role depends on a set of well-defined syntactic rules. This article enumerates those rules using a formal grammar.

Molfile is one of the most widely-used information exchange formats in cheminformatics. Like many formats and toolkits, the V2000 molfile format supports monovalent hydrogen suppression, meaning that hydrogens and bonds to them may exist despite their absence from the molecular graph. This capability raises questions around how exactly suppressed hydrogens should be read and written.

Application Programming Interfaces (APIs) are borne of conflicting demands. On the one hand, clients want an API with a lot of features. On the other hand, too many features can lead to bloat and its malevolent cousins: poor performance; brittleness; and steep learning curves. Good APIs strive for the middle ground, offering just enough functionality to solve important problems, while leaving niche features to extensions.

Open Source Drug Discovery offers an alternative to the proprietary system driving new drug development today. The idea has taken a variety of forms over the years, but in essence is an attempt to replace closed, proprietary processes with open, public ones. A new initiative, the COVID-19 MoonShot, seeks to apply similar principles to development of antiviral treatments for COVID-19. This article offers an overview of the effort.

The lack of effective treatments for COVID-19 has spawned an intense, worldwide drug repurposing effort. Although facing steep odds, this approach offers the advantage that candidates will have already undergone human safety assessment and so could be fast-tracked to approval on that basis. The first of these, an unsuccessful trial of the combination HIV therapy lopinavir-ritonavir was recently published. No doubt there will be many more.

A cheminformatics toolkit or file format is only as good as its molecular representation. The more robust and expressive the representation, the broader the scope of problems that can be addressed. Two frequently-overlooked aspects of molecular representation are formal charge and bond order. These numerical quantities underpin depiction, descriptor calculation, and encoding/decoding among others.

The extensive use of graphs in chemistry to model both reactions and molecules creates challenges for machine learning. Most of the previous and ongoing research in the field solves the problem by translating molecular graphs into forms that fit cleanly within existing paradigms and tools: images; text; and binary fingerprints. Graph representations themselves seem out-of-place at first.