Methods & research
TSR began as a way of describing a protein structure and has been extended to amino acids, drugs, ligands, nucleotides and the interfaces between molecules. Each page below explains one method or analysis and the paper that describes it, and says how to run it with BioTSR where BioTSR implements it.
Every method here builds on the core representation: the TSR method.
Scientific methods
Extensions of the core representation, grouped by what is being described: one molecule, nucleotides, or the interface between two molecules.
One-molecule methods
Amino Acid Grouping TSR
Groups amino acids by structural similarity before generating keys, so that related substructures match even when the residues themselves differ.
Mirror-Image TSR
Extends TSR with keys that distinguish a triangle from its mirror image, so that stereochemistry is captured rather than lost.
Size-Filtering TSR
Compares proteins of very different sizes fairly, by discarding triangles from the larger structure that the smaller one could not possibly contain.
SSE-TSR
Adds secondary structure information to TSR keys, sorting them into 18 groups according to the helix, strand and coil arrangement of each triangle.
Amino Acid TSR
Generates TSR keys at the level of individual amino acids, filtering residues against the expected length for each residue type before triplets are formed.
Nucleotide methods
Nucleotide TSR
Applies the TSR representation to nucleotide residues, filtering the eight RNA and DNA residue types by their expected length before generating keys.
Two-molecule methods
DrugTSR
Represents drug and ligand 3D structures as TSR keys, and builds cross keys spanning both the drug and its target so a complex can be compared as a whole.
Nucleotide–Protein TSR
Builds cross keys from the atoms where a nucleic acid and a protein are in contact, with triangles that span both molecules.
Analysis and research applications
What can be done with keys once they are computed: finding the keys structures share, comparing and clustering structures by their keys, and research models trained on key frequencies. Some of this is built into BioTSR and some is separate research code; each page says which.
Key analysis
Common Keys
Finds the TSR keys that every structure in a set shares, overall and within groups, and the percentage of structures each key occurs in. BioTSR computes common keys from the command line and the percentages from Python.
Visualisation and modelling
Hierarchical Clustering
Clusters structures by TSR key similarity, going from key frequencies through a Jaccard similarity matrix to an average-linkage tree and clustered heat map.
Deep Neural Network
Classifies structures from their TSR key frequencies using a feed-forward network, reporting accuracy, loss and a confusion matrix.
Earlier implementations
The programs the methods were first published with, and the tools written alongside particular papers, are kept unchanged for provenance and for work that depends on them.