Str2Phy
Str2Phy turns a list of protein structures into a structural alignment and a maximum-likelihood tree. Each residue is described by its 3Di state (the local tertiary-interaction environment), the 3Di strings are aligned, and a tree is inferred from the 3Di characters alone, from the amino acids alone, or from both together. The combined analysis is the best-performing approach in the paper cited below.
Citation and scope
Str2Phy supports the following paper. Please cite it if you use the app:
The paper's code is available on GitHub and is exactly reproduced here.
Str2Phy analyses 4 to 53 structures per job, taken from the RCSB PDB and/or the AlphaFold Protein Structure Database (AFDB). Structures are downloaded when the job runs.
Entering structures
Put one structure per line (spaces, commas and semicolons also work). The order of your list is the order of the sequences in the alignment.
- PDB structures:
PDBID_CHAIN, for example1hv4_A. - AlphaFold structures: the UniProt accession, for example
P69905. This is read as the AFDB modelAF-P69905-F1. You can also type the AFDB name yourself, for exampleAF-P69905-F1.
Upper and lower case do not matter (1HV4_a is read as 1hv4_A). Repeated entries are used once.
AlphaFold models and PDB entries can be mixed freely in one job.
The pre-computed example (Figure 6 of the paper)
The paper's ferritin-like analysis (53 structures, amino acids + 3Di, trimmed alignment) has been run in advance and is stored under a fixed Job ID:
On the home page, click See pre-computed results, or type the Job ID into Have a Job ID? and press Open job. It opens the same results page you would get from your own job, so it also shows what every card on that page means (see The Results page). Load example data is a different button: it fills the box with a small set of globins that you can run yourself.
How a job runs
Every job has its own Job ID. Keep it: opening the same address again, even days later, brings you back to the page the job has reached (waiting, choosing, or results).
The Choose page
When the alignment is ready, the page tells you how many structures were used, how long the alignment is (the full alignment, and the alignment after trimming away gappy columns), and the range of chain lengths. Structures that could not be used are listed with the reason, for example a failed download or a chain that is not in the file. Structure details shows, for each structure, its length, missing residues, whether it is an NMR entry (the first model is used), its source and its revision.
Below that are the six analyses of the paper, each as a card:
| Alignment | AA + 3Di (partitioned) | 3Di only | Amino acids only |
|---|---|---|---|
| Trimmed | default · Figure 6 | runs here | runs here |
| Full | download only | runs here | runs here |
Each card shows the model used, the number of distinct site patterns in that alignment, and the estimated running time on the server.
What to select
- AA + 3Di, trimmed is the default, and the analysis behind Figure 6. It uses both kinds of character, each with its own model, on one tree. Start here.
- Trimmed or full. Trimmed drops the columns where most structures have a gap (trimAl, gap threshold 0.35); it is smaller and faster. Full keeps every column. Choose full when you want to check that your result does not depend on the trimming.
- 3Di only and amino acids only use one kind of character on the same alignment. Use them to see how much each contributes, or to compare with the combined tree.
When an analysis is disabled
A card is greyed out, with the reason written on it, when:
- the analysis would take longer than the server allows. The estimate on each card is for an idle server and can be longer when others are running. Large alignments disable the full-length and combined analyses first;
- no columns are left after trimming (this disables every trimmed analysis); or
- the analysis is never run on the server. AA + 3Di, full alignment is always download-only.
If none of the analyses can run, the Run this analysis button is hidden and the page says so.
Downloading instead of running
You do not have to run anything on the server. Download the alignment package gives you the structures,
the amino-acid and 3Di sequences, every alignment, the model files, and a README.md and script
(run_iqtree.sh) that repeat any of the analyses on your own computer, including the ones the server does not
run. You only need IQ-TREE 2.2.2.6 for this; the package README explains where to get it and how to run it on macOS
and Linux.
The Results page
The results page is divided into cards. The same page opens for the pre-computed example.
Run it on your own machine
One download, Download everything, with the alignments, model files, structures, commands, tool versions, checksums and the results of this job. Links beside it give the tree, the consensus tree, the IQ-TREE report and the statistics on their own.
Key numbers
| Tile | Meaning |
|---|---|
| Model | The substitution model(s) used. As in paper means they are the models the paper reports for this analysis. The combined analysis has one model per partition. |
| Log-likelihood | The log-likelihood of the tree (closer to zero is a better fit to the same data). |
| BIC | The Bayesian information criterion; lower is better. Compare it only between analyses of the same alignment. |
| Nodes with UFBoot ≥ 95 | How many internal branches have ultrafast-bootstrap support of 95 or more, out of all internal branches. |
| Columns analysed | Alignment columns used, and the number of distinct site patterns. |
| Tree inference | How long IQ-TREE took to build the tree. |
Phylogenetic tree
An interactive tree you can read without downloading anything.
- Tree: the maximum-likelihood tree, or the consensus tree built from the bootstrap replicates.
- Root: Midpoint (the default), As inferred, or Chosen tip. The tree is inferred unrooted; the root is for display only. Click a tip name to root on it.
- Branches: Proportional to change draws branch lengths; Aligned tips lines the tips up.
- Support: coloured dots (blue: UFBoot ≥ 95; amber: 80 to 95; open: below 80), UFBoot numbers, SH-aLRT / UFBoot, or none.
- Find highlights a structure by name. SVG and PNG save the picture.
Side cards
- Model and fit: the analysis, its model, the log-likelihood and BIC, and the seed.
- Branch support: the number of branches with UFBoot ≥ 95 and SH-aLRT ≥ 80, and the mean, median and lowest support (1000 ultrafast-bootstrap replicates and 1000 SH-aLRT replicates).
- How this was run: the exact IQ-TREE command, with a copy button, and the versions of the tools used.
Alignment and structures
- Alignment: the number of structures, the full and trimmed lengths, and the number of distinct site patterns. For the combined analysis a bar shows the two partitions (amino acids, then 3Di).
- Structures: every structure in the job with its length, missing residues, NMR models and revision. Type in the filter to find one, and click a row to highlight it in the tree.
Files
Everything is downloadable, grouped as:
- Results: tree with support values (Newick), consensus tree (Newick), IQ-TREE report, IQ-TREE log, statistics (JSON) and the pipeline log;
- Alignments: amino acid, 3Di and combined, full and trimmed (FASTA); the ones used for this tree are marked;
- Models and partitions: partition files and the 3Di model (NEXUS and plain text);
- Everything: the results package, the alignment package (without the tree), and JSON summaries of the alignment and of the structures.
Using the code without the app
- To repeat a tree from a Str2Phy download, all you need is IQ-TREE (version 2.2.2.6). The package
contains the alignments and everything else; its
README.mdandrun_iqtree.shdo the rest. - To reproduce everything from the structures, the alignment is built with three tools, and the tree with IQ-TREE. Follow the instructions in the paper's GitHub repository.
Contact
Developed and maintained by Ashar Malik
(ashar.malik@uq.edu.au)
BioSig Lab · School of Chemistry and Molecular Biosciences · The University of Queensland