Julia Haag’s research while affiliated with Heidelberg Institute for Theoretical Studies and other places

Maximum Likelihood (ML) based phylogenetic inference is time- and resource-intensive, especially when initiating multiple independent inferences from distinct comprehensive tree topologies. Performing multiple independent inferences is often required to (sufficiently) explore the vast search space of possible unrooted binary tree topologies. Yet, these independent inferences do not necessarily converge to a single phylogeny or at least topologically highly similar trees. While foreasy-to-analyze multiple sequence alignments (MSAs), one is likely to obtain a conclusive, single phylogeny, difficult-to-analyze MSAs yield topologically highly distinct, yet statistically indistinguishable tree topologies. In 2022, we proposed a compute-intensive approach to quantify the inherent difficulty of a phylogenetic analysis for a specific, given MSA, and also trained a machine-learning based prediction model called Pythia to substantially reduce the computational cost of determining the difficulty. Pythia can predict the difficulty for a given MSA with high accuracy, while being substantially faster than even a single ML tree inference. Pythia predicts the difficulty on a scale from 0 (easy) to 1 (difficult). Here, we present all improvements to Pythia that we have introduced since our initial publication in 2022. We trained a new prediction model using approximately three times more MSAs and a new type of machine learning model. We improved the runtime of two feature computations, and we also introduced two additional prediction features. Our latest version Pythia 2.0 is slightly more accurate than our initial version and is also approximately twice as fast. Finally, we also present and make available, the novel and easy-to-use command line tool PyDLG that allows to compute the ground-truth difficulty seamlessly for a given MSA. This ground-truth difficulty can be used, for instance, as a prediction target for training a new Pythia model.

Motivation Genotype datasets typically contain a large number of single nucleotide polymorphisms for a comparatively small number of individuals. To identify similarities between individuals and to infer an individual’s origin or membership to a population, dimensionality reduction techniques are routinely deployed. However, inherent (technical) difficulties such as missing or noisy data need to be accounted for when analyzing a lower dimensional representation of genotype data, and the intrinsic uncertainty of such analyses should be reported in all studies. However, to date, there exists no stability assessment technique for genotype data that can estimate this uncertainty. Results Here, we present Pandora, a stability estimation framework for genotype data based on bootstrapping. Pandora computes an overall score to quantify the stability of the entire embedding, infers per-individual support values, and also deploys a k-means clustering approach to assess the uncertainty of assignments to potential cultural groups. Using published empirical and simulated datasets, we demonstrate the usage and utility of Pandora for studies that rely on dimensionality reduction techniques. Availability and Implementation Pandora is available on GitHub https://github.com/tschuelia/Pandora. Supplementary information Supplementary data are available online.

Estimating the statistical robustness of the inferred tree(s) constitutes an integral part of most phylogenetic analyses. Commonly, one computes and assigns a branch support value to each inner branch of the inferred phylogeny. The still most widely used method for calculating branch support on trees inferred under maximum likelihood (ML) is the Standard, nonparametric Felsenstein bootstrap support (SBS). Due to the high computational cost of the SBS, a plethora of methods has been developed to approximate it, for instance, via the rapid bootstrap (RB) algorithm. There have also been attempts to devise faster, alternative support measures, such as the SH-aLRT (Shimodaira–Hasegawa-like approximate likelihood ratio test) or the UltraFast bootstrap 2 (UFBoot2) method. Those faster alternatives exhibit some limitations, such as the need to assess model violations (UFBoot2) or unstable behavior in the low support interval range (SH-aLRT). Here, we present the educated bootstrap guesser (EBG), a machine learning-based tool that predicts SBS branch support values for a given input phylogeny. EBG is on average 9.4 (σ=5.5) times faster than UFBoot2. EBG-based SBS estimates exhibit a median absolute error of 5 when predicting SBS values between 0 and 100. Furthermore, EBG also provides uncertainty measures for all per-branch SBS predictions and thereby allows for a more rigorous and careful interpretation. EBG can, for instance, predict SBS support values on a phylogeny comprising 1,654 SARS-CoV2 genome sequences within 3 h on a mid-class laptop. EBG is available under GNU GPL3.

Despite tremendous efforts in the past decades, relationships among main avian lineages remain heavily debated without a clear resolution. Discrepancies have been attributed to diversity of species sampled, phylogenetic method and the choice of genomic regions1–3. Here we address these issues by analysing the genomes of 363 bird species⁴ (218 taxonomic families, 92% of total). Using intergenic regions and coalescent methods, we present a well-supported tree but also a marked degree of discordance. The tree confirms that Neoaves experienced rapid radiation at or near the Cretaceous–Palaeogene boundary. Sufficient loci rather than extensive taxon sampling were more effective in resolving difficult nodes. Remaining recalcitrant nodes involve species that are a challenge to model due to either extreme DNA composition, variable substitution rates, incomplete lineage sorting or complex evolutionary events such as ancient hybridization. Assessment of the effects of different genomic partitions showed high heterogeneity across the genome. We discovered sharp increases in effective population size, substitution rates and relative brain size following the Cretaceous–Palaeogene extinction event, supporting the hypothesis that emerging ecological opportunities catalysed the diversification of modern birds. The resulting phylogenetic estimate offers fresh insights into the rapid radiation of modern birds and provides a taxon-rich backbone tree for future comparative studies.

Motivation Genotype datasets typically contain a large number of single nucleotide polymorphisms for a comparatively small number of individuals. To identify similarities between individuals and to infer an individual’s origin or membership to a cultural group, dimensionality reduction techniques are routinely deployed. However, inherent (technical) difficulties such as missing or noisy data need to be accounted for when analyzing a lower dimensional representation of genotype data, and the uncertainty of such an analysis should be reported in all studies. However, to date, there exists no stability estimation technique for genotype data that can estimate this uncertainty. Results Here, we present Pandora, a stability estimation framework for genotype data based on bootstrapping. Pandora computes an overall score to quantify the stability of the entire embedding, perindividual support values, and deploys a k -means clustering approach to assess the uncertainty of assignments to potential cultural groups. In addition to this bootstrap-based stability estimation, Pandora offers a sliding-window stability estimation for whole-genome data. Using published empirical and simulated datasets, we demonstrate the usage and utility of Pandora for studies that rely on dimensionality reduction techniques. Data and Code Availability Pandora is available on GitHub https://github.com/tschuelia/Pandora . All Python scripts and data to reproduce our results are available on GitHub https://github.com/tschuelia/PandoraPaper . Contact julia.haag@h-its.org

Estimating the statistical robustness of the inferred tree(s) constitutes an integral part of most phylogenetic analyses. Commonly, one computes and assigns a branch support value to each inner branch of the inferred phylogeny. The most widely used method for calculating branch support on trees inferred under Maximum Likelihood (ML) is the Standard, non-parametric Felsenstein Bootstrap Support (SBS). Due to the high computational cost of the SBS, a plethora of methods has been developed to approximate it, for instance, via the Rapid Bootstrap (RB) algorithm. There have also been attempts to devise faster, alternative support measures, such as the SH-aLRT (Shimodaira–Hasegawalike approximate Likelihood Ratio Test) or the UltraFast Bootstrap 2 (UFBoot2) method. Those faster alternatives exhibit some limitations, such as the need to assess model violations (UFBoot2) or meaningless low branch support intervals (SH-aLRT). Here, we present the Educated Bootstrap Guesser (EBG), a machine learning-based tool that predicts SBS branch support values for a given input phylogeny. EBG is on average 9.4 ( σ = 5.5) times faster than UFBoot2. EBG-based SBS estimates exhibit a median absolute error of 5 when predicting SBS values between 0 and 100. Furthermore, EBG also provides uncertainty measures for all per-branch SBS predictions and thereby allows for a more rigorous and careful interpretation. EBG can predict SBS support values on a phylogeny comprising 1654 SARS-CoV2 genome sequences within 3 hours on a mid-class laptop. EBG is available under GNU GPL3. Data and Code Availability github.com/wiegertj/EBG github.com/wiegertj/EBG-train Contact julius-wiegert@web.de

Motivation Simulating Multiple Sequence Alignments (MSAs) using probabilistic models of sequence evolution plays an important role in the evaluation of phylogenetic inference tools, and is crucial to the development of novel learning-based approaches for phylogenetic reconstruction, for instance, neural networks. These models and the resulting simulated data need to be as realistic as possible to be indicative of the performance of the developed tools on empirical data and to ensure that neural networks trained on simulations perform well on empirical data. Over the years, numerous models of evolution have been published with the goal to represent as faithfully as possible the sequence evolution process and thus simulate empirical-like data. In this study, we simulated DNA and protein MSAs under increasingly complex models of evolution with and without insertion/deletion (indel) events using a state-of-the-art sequence simulator. We assessed their realism by quantifying how accurately supervised learning methods are able to predict whether a given MSA is simulated or empirical. Results Our results show that we can distinguish between empirical and simulated MSAs with high accuracy using two distinct and independently developed classification approaches across all tested models of sequence evolution. Our findings suggest that the current state-of-the-art models fail to accurately replicate several aspects of empirical MSAs, including site-wise rates as well as amino acid and nucleotide composition. Data and Code Availability All simulated and empirical MSAs, as well as all analysis results, are available at https://cme.h-its.org/exelixis/material/simulation_study.tar.gz. All scripts required to reproduce our results are available at https://github.com/tschuelia/SimulationStudy and https://github.com/JohannaTrost/seqsharp.

Phylogenetic inferences under the Maximum-Likelihood criterion deploy heuristic tree search strategies to explore the vast search space. Depending on the input dataset, searches from different starting trees might all converge to a single tree topology. Often, though, distinct searches infer multiple topologies with large log-likelihood score differences or yield topologically highly distinct, yet almost equally likely, trees. Recently, Haag et al. introduced an approach to quantify, and implemented machine learning methods to predict, the dataset difficulty with respect to phylogenetic inference. Easy MSAs exhibit a single likelihood peak on their likelihood surface, associated with a single tree topology to which most, if not all, independent searches rapidly converge. As difficulty increases, multiple locally optimal likelihood peaks emerge, yet from highly distinct topologies. To make use of this information, we introduce and implement an adaptive tree search heuristic in RAxML-NG, which modifies the thoroughness of the tree search strategy as a function of the predicted difficulty. Our adaptive strategy is based upon three observations. First, on easy datasets, searches converge rapidly and can hence be terminated at an earlier stage. Second, over-analyzing difficult datasets is hopeless, and thus it suffices to quickly infer only one of the numerous almost equally likely topologies to reduce overall execution time. Third, more extensive searches are justified and required on datasets with intermediate difficulty. While the likelihood surface exhibits multiple locally optimal peaks in this case, a small proportion of them is significantly better. Our experimental results for the adaptive heuristic on 9,515 empirical and 5,000 simulated datasets with varying difficulty exhibit substantial speedups, especially on easy and difficult datasets (53% of total MSAs), where we observe average speedups of more than 10x. Further, approximately 94% of the inferred trees using the adaptive strategy are statistically indistinguishable from the trees inferred under the standard strategy (RAxML-NG).

Maximum likelihood (ML) is a widely used phylogenetic inference method. ML implementations heavily rely on numerical optimization routines that use internal numerical thresholds to determine convergence. We systematically analyze the impact of these threshold settings on the log-likelihood and runtimes for ML tree inferences with RAxML-NG, IQ-TREE, and FastTree on empirical datasets. We provide empirical evidence that we can substantially accelerate tree inferences with RAxML-NG and IQ-TREE by changing the default values of two such numerical thresholds. At the same time, altering these settings does not significantly impact the quality of the inferred trees. We further show that increasing both thresholds accelerates the RAxML-NG bootstrap without influencing the resulting support values. For RAxML-NG, increasing the likelihood thresholds ϵLnL and ϵbrlen to 10 and 103, respectively, results in an average tree inference speedup of 1.9 ± 0.6 on Data collection 1, 1.8 ± 1.1 on Data collection 2, and 1.9 ± 0.8 on Data collection 2 for the RAxML-NG bootstrap compared to the runtime under the current default setting. Increasing the likelihood threshold ϵLnL to 10 in IQ-TREE results in an average tree inference speedup of 1.3 ± 0.4 on Data collection 1 and 1.3 ± 0.9 on Data collection 2. Availability and implementation All MSAs we used for our analyses, as well as all results, are available for download at https://cme.h-its.org/exelixis/material/freeLunch_data.tar.gz. Our data generation scripts are available at https://github.com/tschuelia/ml-numerical-analysis.

Motivation: Simulating sequence evolution plays an important role in the development and evaluation of phylogenetic inference tools. Naturally, the simulated data needs to be as realistic as possible to be indicative of the performance of the developed tools on empirical data. Over the years, numerous phylogenetic sequence simulators, employing various models of evolution, have been published with the goal to simulate such empirical-like data. In this study, we simulated DNA and protein Multiple Sequence Alignments (MSAs) under increasingly complex models of evolution with and without insertion/deletion (indel) events using a state-of-the-art sequence simulator. We assessed their realism by quantifying how well supervised learning methods are able to predict whether a given MSA is simulated or empirical. Results: Our results show that we can distinguish between empirical and simulated MSAs with high accuracy using two distinct and independently developed classification approaches across all tested models of sequence evolution. Our findings suggest that the current state-of-the-art models fail to accurately replicate the process of evolution.