Large-Scale Multiple Sequence Alignment Tandy Warnow Founder Professor of Computer Science The University of Illinois at Urbana-Champaign http://tandy.cs.illinois.edu

1kp: Thousand Transcriptome Project G. Ka-Shu Wong U Alberta J. Leebens-Mack U Georgia N. Wickett Northwestern N. Matasci iPlant T. Warnow, S. Mirarab, N. Nguyen UIUC UCSD UCSD Plus many many other people… Plant Tree of Life based on transcriptomes of ~1200 species More than 13,000 gene families (most not single copy) Gene Tree Incongruence Challenge: Alignments and trees on > 100,000 sequences

1. 2 classes of MC: easy, moderate-to-difficult 2. true alignment 3. 2 classes: ClustalW, everything else Alignment error, measured this way, isn't a perfect predictor of tree error, measured this way. 1000-taxon models, ordered by difficulty (Liu et al., Science 324(5934):1561-1564, 2009) 3

Estimate ML tree on merged alignment Re-aligning on a tree A B D C A B Decompose dataset C D Align subsets A B Comment on subset size C D Estimate ML tree on merged alignment ABCD Merge sub-alignments

SATé and PASTA Algorithms Tree Obtain initial alignment and estimated ML tree Estimate ML tree on new alignment Use tree to compute new alignment Alignment Repeat until termination condition, and return the alignment/tree pair with the best ML score

1000 taxon models, ordered by difficulty, Liu et al 1000 taxon models, ordered by difficulty, Liu et al., Science 324(5934):1561-1564, 2009 For moderate-to-difficult datasets, SATe gets better trees and alignments than all other estimated methods. Close to what you might get if you had access to true alignment. Opens up a new realm of possibility: Datasets currently considered “unalignable” can in fact be aligned reasonably well. This opens up the feasibility of accurate estimations of deep evolutionary histories using a wider range of markers. TRANSITION: can we do better? What about smaller simulated datasets? And what about biological datasets? 24 hour SATé-I analysis, on desktop machines (Similar improvements for biological datasets)

1000-taxon models ranked by difficulty SATé-2 better than SATé-1 1000-taxon models ranked by difficulty SATé-1 (Liu et al., Science 2009): can analyze up to 8K sequences SATé-2 (Liu et al., Systematic Biology 2012): can analyze up to ~50K sequences

PASTA: even better than SATé-2 PASTA: Mirarab, Nguyen, and Warnow, J Comp. Biol. 2015 Simulated RNASim datasets from 10K to 200K taxa Limited to 24 hours using 12 CPUs Not all methods could run (missing bars could not finish)

1kp: Thousand Transcriptome Project G. Ka-Shu Wong U Alberta J. Leebens-Mack U Georgia N. Wickett Northwestern N. Matasci iPlant T. Warnow, S. Mirarab, N. Nguyen UIUC UCSD UCSD Plus many many other people… Plant Tree of Life based on transcriptomes of ~1200 species More than 13,000 gene families (most not single copy) Gene Tree Incongruence Challenge: Alignments and trees on > 100,000 sequences

1KP dataset: more than 100,000 p450 amino-acid sequences, many fragmentary

datasets with fragments. 1KP dataset: more than 100,000 p450 amino-acid sequences, many fragmentary All standard multiple sequence alignment methods we tested performed poorly on datasets with fragments.

UPP UPP = “Ultra-large multiple sequence alignment using Phylogeny-aware Profiles” Nguyen, Mirarab, and Warnow. Genome Biology, 2014. Purpose: highly accurate large-scale multiple sequence alignments, even in the presence of fragmentary sequences.

UPP Uses an ensemble of HMMs UPP = “Ultra-large multiple sequence alignment using Phylogeny-aware Profiles” Nguyen, Mirarab, and Warnow. Genome Biology, 2014. Purpose: highly accurate large-scale multiple sequence alignments, even in the presence of fragmentary sequences. Uses an ensemble of HMMs

Simple idea (not UPP) Select random subset of sequences, and build “backbone alignment” Construct a Hidden Markov Model (HMM) on the backbone alignment Add all remaining sequences to the backbone alignment using the HMM

One Hidden Markov Model for the entire alignment?

One Hidden Markov Model for the backbone alignment? HMM 1

Or 2 HMMs? HMM 1 HMM 2

Or 4 HMMs? HMM 1 HMM 2 the bit score doesn’t depend on the size of the sequence database, only on the profile HMM and the target sequence HMM 3 HMM 4

Or all 7 HMMs? HMM 1 HMM 2 m HMM 4 HMM 7 HMM 3 HMM 5 HMM 6 the bit score doesn’t depend on the size of the sequence database, only on the profile HMM and the target sequence HMM 3 HMM 5 HMM 6

UPP Algorithmic Approach Select random subset of full-length sequences, and build “backbone alignment” with PASTA Construct an “Ensemble of Hidden Markov Models” on the backbone alignment Add all remaining sequences to the backbone alignment using the Ensemble of HMMs

Evaluation Simulated datasets (some have fragmentary sequences): 10K to 1,000,000 sequences in RNASim – complex RNA sequence evolution simulation 1000-sequence nucleotide datasets from SATé papers 5000-sequence AA datasets (from FastTree paper) 10,000-sequence Indelible nucleotide simulation Biological datasets: Proteins: largest BaliBASE and HomFam RNA: 3 CRW datasets up to 28,000 sequences

RNASim Million Sequences: alignment error Notes: We show alignment error using average of SP-FN and SP- FP.  UPP variants have better alignment scores than PASTA.  (Not shown: Total Column Scores – PASTA more accurate than UPP)  No other methods tested could complete on these data

RNASim Million Sequences: tree error Using 12 processors: UPP(Fast,NoDecomp) took 2.2 days, UPP(Fast) took 11.9 days, and PASTA took 10.3 days

UPP is very robust to fragmentary sequences Under high rates of evolution, PASTA is badly impacted by fragmentary sequences (the same is true for other methods). UPP continues to have good accuracy even on datasets with many fragments under all rates of evolution. Performance on fragmentary datasets of the 1000M2 model condition

UPP Running Time Wall-clock time used (in hours) given 12 processors

What about BAli-Phy?  BAli-Phy (Redelings and Suchard): leading method for statistical co-estimation of alignments and trees: Like Bayesian phylogeny estimation, it is expected to be the most rigorous and accurate technique for estimating trees and alignments!

BAli-Phy: Better than PASTA! Alignment Accuracy (TC score) MAFFT PASTA BAli-Phy 40% 30% 20% 10% 0% # Taxa: Total-Column Score 100 Indelible (DNA) 200 100 RNAsim (RNA) 200 Simulator Simulated datasets with 100 or 200 sequences. *Averages over 10 replicates

But: BAli-Phy is limited to small datasets From www.bali-phy.org/README.html, 5.2.1. Too many taxa? “BAli-Phy is quite CPU intensive, and so we recommend using 50 or fewer taxa in order to limit the *me required to accumulate enough MCMC samples. (Despite this recommendation, data sets with more than 100 taxa have occasionally been known to converge.) We recommend initially pruning as many taxa as possible from your data set, then adding some back if the MCMC is not too slow.”

Estimate ML tree on merged alignment Re-aligning on a tree A B D C A B Decompose dataset C D Align subsets: MAFFT A B Comment on subset size C D Estimate ML tree on merged alignment ABCD Merge sub-alignments

Re-aligning on a tree Decompose dataset Align subsets: BAli-Phy?? Comment on subset size C D Estimate ML tree on merged alignment ABCD Merge sub-alignments

Comparing default PASTA to PASTA+BAli-Phy on simulated datasets with 1000 sequences Decomposition to 100-sequence subsets, one iteration of PASTA+BAli-Phy

Results on 10,000-sequence datasets, backbone size 1000 Comparing UPP variants where the backbone alignment is computed using either default PASTA or PASTA+BAli-Phy Total Column Score 0.100 PAST A+BAli−Phy Better 0.075 PASTA+BAli−Phy ● data 0.050 ● ● Indelible M2 RNAsim ● ● ● ●● ● ● PASTA Better 0.025 ● 0.000 0.000 0.025 0.050 PASTA 0.075 0.100 Recall (SP−Score) Tree Error: Delta RF (FastTree−2) 1.000 PAST A+BAli−Phy Better 0.015 PASTA+BAli−Phy Bette ●r 0.975 ● PASTA+BAli−Phy ●● 0.010 ● ● ● ● ●● PASTA 0.950 ● ● ● ● ● ● ● PASTA Better 0.005 ● ● ● PASTA Better 0.925 0.900 0.000 0.900 0.925 0.950 PASTA 0.975 1.000 0.000 0.005 0.010 PASTA+BAli−Phy 0.015 Results on 10,000-sequence datasets, backbone size 1000

Summary Large-scale multiple sequence alignment (MSA) is achievable with good accuracy using divide-and-conquer plus iteration, and enable preferred MSA methods to be used on very large datasets PASTA and UPP can provide good accuracy on large (million- sequence) datasets with high heterogeneity Fragmentary sequences present additional challenges that UPP can address PASTA and UPP at https://github.com/smirarab/ PASTA+BAli-Phy at http://github.com/MGNute/pasta

Acknowledgments PASTA and UPP: Nam Nguyen (now postdoc at UIUC) and Siavash Mirarab (now faculty at UCSD), undergrad: Keerthana Kumar (at UT-Austin) PASTA+BAli-Phy: Mike Nute (PhD student at UIUC) Current NSF grants: ABI-1458652 (multiple sequence alignment) Grainger Foundation (at UIUC), and UIUC TACC, UTCS, Blue Waters, and UIUC campus cluster PASTA, UPP, SEPP, and TIPP are available on github at https://github.com/smirarab/; see also PASTA+BAli-Phy at http://github.com/MGNute/pasta