Slides for my contribution to the symposium "Beyond the Tree of Life", held 16-18 March 2016, at the Trippenhuis in Amsterdam, the Netherlands.

1. FROM PUBLIC DNA SEQUENCE
DATA
PITFALLS, CHALLENGES AND
SOLUTIONS
Rutger Vos, Naturalis Biodiversity Center
Twitter: @rvosa

2. Public DNA sequence data. A
lot.

4. Want a moonshot? Build a
rocket.

5. Assembling a Tree of Life
If you want to build The
Tree of Life you will
need to build a system
for building Trees of
Life.
Such a system has
many moving parts:
-data mining
-marker selection
-tree inference
-tree calibration
-tree grafting

6. Data mining
Keyword searches allow one to locate specific
genes for specific taxa, assuming that gene
names are standardized and applied correctly.

7. Data mining
Similarity ("BLAST") searches allow one to
locate sequences that are similar to a query
sequence

8. Pitfalls in data mining
 Bad gene naming conventions
 Incomplete knowledge of
orthology
 Ambiguous taxonomy

9. Possible solutions
 TNRS helps resolve ambiguities in names
(e.g. synonyms): http://taxosaurus.org
 Sequence clustering databases help avoid
keyword searching and haphazard BLASTing,
e.g.:
http://phylota.net

11. Digression: orthology
assignment
 Based on gene names?
Hopeless: outside of very few markers, gene
names are a mess
 Based on pairwise genome comparisons?
Sort-of done for proteins (e.g. InParanoid) but
not for non-coding
 Based on pairwise reciprocal best BLAST
hits?
Quick and cheap, but error prone depending
on losses and gains, and database
completeness

12. Supermatrix packing
How to
optimize
combinations
of markers to
maximize taxon
sampling and
minimize
sparseness?

13. Tree inference
 The current gold standard in species trees
from multilocus alignments: *BEAST. Very
expensive.
 Cheaper, scalable Bayesian tree inference is
provided by other tools, e.g. ExaBayes.
 Maybe the two can be combined? For
example:
 A cheap, large, taxonomically broad backbone
 Within-clade relationships resolved more
expensively

14. Tree calibration
 Fossils now available
through web service of
http://fossilcalibrations.or
g
 Scalable tree calibration
tools now exist, e.g.
treePL
 *BEAST has more
sophisticated methods

15. Tree grafting
Pitfalls:
• What if your exemplar species aren't on either side of the root?
• Grafting could then easily result in negative branch lengths
• Also, node density effects will lead to younger nodes in the backbone

16. All in one: SUPERSMART
The SUPERSMART
platform helps
assemble high quality
marker sets and
analyze them using
configurable divide-and-
conquer tree inference
approaches.
The platform is
available as a VM, a
Docker container, or as
an easy to install (using
Puppet)software stack.

17. Results: Primates and Palms

18. Conclusions
 More and more moving parts for a ToL
moonshot are becoming available
 Workflows that result in high-quality estimates
can be composed from these
 SUPERSMART uses a recursive workflow with
different inference methods at different levels

19. Acknowledgements
 Alexandre Antonelli
 Hannes Hettling
 Mike Sanderson
 Bengt Oxelman
 Karin Nilsson
 Mats Töpel
 Hervé Sauquet
 Henrik Nilsson
 Daniele Silvestro
 Fabien Condamine
 Ruud Scharn

20. Questions?
 Thanks for listening!
 Contact me at @rvosa
 For more about our project:
 http://www.supersmart-project.org