phylogenetic placement - iqtree/iqtree2 GitHub Wiki
This recipe shows you how to add new sequences to an existing tree (phylogenetic placement), then to perform a global re-optimisation of that tree.
Sometimes you have an existing phylogeny, and you'd like to add new sequences it without altering the existing relationships. This is increasingly common in what's called 'online phylogenetics'. For example, in an ongoing pathogen outbreak we might get a lot of new genomes arriving in our database every day. Rather than infer a new tree from scratch each day, it's better to start with yesterday's tree and just add the new genomes to it. This preprint, shows some recent work on online phylogenetics, and compares a lot of different approaches. In this recipe I'll cover how to do this in a Maximum Likelihood framework using IQ-TREE.
If you have very large datasets, i.e. more than many tens of thousands of sequences, you should consider using UShER & matOptimize, or MAPLE for your online phylogenetics analyses. These pieces of software can comfortably handle alignments of millions of sequences, with tens of thousands of new sequences being added every day. There are tradeoffs (e.g. matOptimize uses parsimony) but they are often perfectly acceptable (that's covered in the preprint linked to above).
For this example, I'll imagine you have an existing tree and alignment of 17 sequences (T1 to T17). Since estimating your tree, you've added three new sequences to your alignment (NEW_1, NEW_2, and NEW_3). Now you want to add those new sequences to your existing tree.
-
sequences.fa: Your alignment of 20 sequences (T1toT17, andNEW_1toNEW_3) -
existing_tree.nex: An existing tree of 17 sequences (T1 to T17)
iqtree -s sequences.fa -g existing_tree.nex
-
-gsets a constraint tree with our existing 17 sequences -
-spasses the alignment of the existing 17 sequences and the 3 new sequences
Your output tree will now look like this:
/--+ T2
|
| /--------+ T3
| /+
| |\-------------+ NEW 2
| |
| /---------+ /--+ T9
| | | /--+
| | \------------------------+ \--+ NEW 3
+----+ |
| | \-------+ T11
| |
| \---------------------+ T4
|
=+ /-+ T5
| |
| /---------------+ /----------------------+ T17
| | \-------------------------------+
| /----------------------+ \---+ NEW 1
| | |
| | | /------------+ T13
| | \-----+
| /-+ \--------+ T15
| | |
| | | /----+ T6
| | |/-----+
| | \+ \----+ T10
\---------------------------------------+ |
| \----+ T7
|
| /-----------------+ T8
| |
\---+ /-------+ T12
| |
\-----+ /--------------+ T14
| |
\------------------------+ /-+ T1
\------+
\-----------+ T16
|---------------------------------|--------------------------------|---------------------------------|--------------------------------|--
0 0.25 0.5 0.75 1
substitutions/site
Our sequence alignment contains 20 sequences: T1 to T17, and NEW_1 to NEW_3. The sequences T1 to T17 all appear in our tree existing_tree.nex. The option -g uses existing_tree.nex as a constraint tree, which means that the topology of T1 to T17 will be fixed for this analysis. The sequences NEW_1, NEW_2, and NEW_3 are not in the constraint tree. So, IQ-TREE will initially place these sequences onto the constraint tree using Maximum Parsimony, and then it will do standard ML optimisation of the tree without altering the underlying constraint tree. In other words, it will use the standard IQ-TREE search algorithm to try moving NEW_1, NEW_2, and NEW_3 around on the constraint tree, and will keep these changes if they improve the likelihood. This optimisation helps for two reasons. First, placing sequences with sequential parsimony placement doesn't always give you the best parsimony placement, and second, even the best parsimony placement can differ from the best placement under a full Likelihood model.
In some cases, e.g. if you've added a lot of new sequences to an existing tree, you might want to then re-optimise the tree without the constraints. This is typically what we do for most online phylogenetics applications, for example. To do this, you would use the following options:
-
-tto set the starting tree -
-sto pass the alignment
i.e.
iqtree -s sequences.fa -t sequences.treefile -pre reoptimised
This commandline simply uses the tree output from the first commandline (sequences.treefile) as the starting tree for a new analysis. That way you start from the best estimate of the tree you have, and IQ-TREE will then try to further optimise it in the absence of constraints.
Here's some code to build this example yourself from scratch. Note that this assumes you're working on Unix, and you'll need newick_utils for this to work.
# simulate an alignment of 20 sequences (set the seed for reproducibility only)
iqtree2 --alisim sequences -t RANDOM{yh/20} -seed 438579088 -af fasta -m JC
# check the names
grep '>' sequences.fa
# replace names of 3 of them to be 'NEW_N'
sed -i -e 's/T18/NEW_1/g' sequences.fa
sed -i -e 's/T19/NEW_2/g' sequences.fa
sed -i -e 's/T20/NEW_3/g' sequences.fa
# check the names again
grep '>' sequences.fa
# prune the NEW_N sequences from the tree using newick_utils
nw_prune sequences.treefile T18 T19 T20 > existing_tree.nex
# check that you have the files you need
ls -lh
# run IQ-TREE to add NEW_N seqs back
iqtree -g existing_tree.nex -s sequences.fa -pre seqs_added -seed 438579088
# optional: do a global re-optimisation of the tree, using previous tree as start
iqtree -s sequences.fa -t seqs_added.treefile -pre re_optimised -seed 438579088
Last tested with IQ-TREE 2.2.0.3