Tip rate estimates can predict future diversification, but are unreliable and context dependent

Tip rate estimates can predict future diversification, but are unreliable and context dependent | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results Tip rate estimates can predict future diversification, but are unreliable and context dependent View ORCID Profile Grace I Ridder , View ORCID Profile Jan Smyčka , View ORCID Profile David Storch , View ORCID Profile Arne Ø Mooers , View ORCID Profile Sarah P Otto doi: https://doi.org/10.1101/2025.10.06.680809 Grace I Ridder 1 Center for Theoretical Study, Charles University and Czech Academy of Sciences , Prague, Czechia 2 Department of Ecology, Faculty of Science, Charles University , Prague, Czechia Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Grace I Ridder Jan Smyčka 1 Center for Theoretical Study, Charles University and Czech Academy of Sciences , Prague, Czechia 3 Department of Biological Sciences, Simon Fraser University , Burnaby, British Columbia, Canada 4 Department of Zoology and Biodiversity Research Centre, University of British Columbia , Vancouver, British Columbia, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Jan Smyčka For correspondence: smyckaj{at}gmail.com David Storch 1 Center for Theoretical Study, Charles University and Czech Academy of Sciences , Prague, Czechia 2 Department of Ecology, Faculty of Science, Charles University , Prague, Czechia Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for David Storch Arne Ø Mooers 3 Department of Biological Sciences, Simon Fraser University , Burnaby, British Columbia, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Arne Ø Mooers Sarah P Otto 4 Department of Zoology and Biodiversity Research Centre, University of British Columbia , Vancouver, British Columbia, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Sarah P Otto Abstract Full Text Info/History Metrics Data/Code Preview PDF Abstract Understanding the variability of processes leading to the emergence of new lineages is one of the major tasks of macroevolution as a scientific field. Recent years have seen the rise of rate-variable diversification models and metrics that estimate the rates of species diversification at the tips of phylogenetic trees and are thus potentially useful for predicting future evolutionary success of individual species. These methods use various assumptions about the variability and heritability of diversification rates. However, the general performance of rate-variable diversification methods have never been consistently tested against real world data. Here we explore the capacity of multiple rate-variable diversification methods to predict near-future diversification using temporal slices of empirical fossil and extant phylogenies. We do this using a newly developed approach similar to generalized linear models, allowing us to quantify the relationship between predictor tip rates and subsequent diversification rates derived from a probability distribution of numbers of daughter species. We find that tip rates estimated from current methods have non-zero but limited capacity to predict diversification in both fossil and extant phylogenies. The quality of the predictions depends not only on the methods used but also on the specific phylogeny, suggesting that diversification dynamics in some taxa may be more predictable in principle. Our results suggest that future cladogenesis can be, to a certain extent, predicted using existing tip rate methods, but the quality of such predictions is highly variable and depends on factors that are difficult to evaluate in practical applications. Introduction Species diversification, the process of formation of biological diversity through speciation and extinction, runs at different paces in different evolutionary lineages ( Simpson 1944 ; Gould and Eldredge 1977 ; Jablonski 2008 ). Understanding this variation is crucial for inferring processes of the evolutionary past, explaining current patterns of species diversity, and potentially for predicting the fate of biodiversity in the future. Over the last several decades, researchers have developed an impressive toolbox of methods for capturing and exploring the variability of species diversification across time calibrated phylogenetic trees ( Jetz et al. 2012 ; Rabosky 2014 ; Maliet et al. 2019 ). These advances have been pivotal for associating different diversification rates with particular lineage characteristics, e.g. to identify links between species diversification and the appearance of key evolutionary innovations ( Rabosky 2014 ), various traits ( Helmstetter et al. 2023 ) or geographic features ( Machac 2020 ; Rabosky 2022 ; Smyčka et al. 2023 ). Similarly, there have been efforts, both qualitative and quantitative, to extrapolate our estimates of species diversification into the future ( Rosenzweig 2001 ; Storch et al. 2022 ) and use them for evaluating lineages for conservation planning ( Kling et al. 2018 ; Cantalapiedra et al. 2019 ). Despite the wide usage of these diversification methods, there is an ongoing debate about the appropriateness and mutual compatibility of the assumptions they make about the diversification process ( Maliet et al. 2019 ; Title and Rabosky 2019 ; Ronquist et al. 2021 ). For example, three methods commonly used for estimating rates of species diversification at the phylogeny tips (so called tip rates), all use dramatically different assumptions about the mechanisms of diversification variation among the lineages. A widely used model for predicting tip rates called BAMM (Rabosky 2014) assumes that diversification rates occasionally abruptly change through evolutionary time, and the rest of the variation in evolutionary rates is less pronounced and gradual ( Fig 1a ). An alternative approach referred to as ClaDS ( Maliet et al. 2019 ) assumes that the diversification rates change at every cladogenetic event but are still auto-correlated across the tree ( Fig 1b ). The commonly used model-free DR metric ( Jetz et al. 2012 ) posits that species diversification rates can be approximated by an inverted value of species isolation progressively down-weighted by distance from each focal tip ( Redding and Mooers 2006 ; Fig 1c ). Download figure Open in new tab Figure 1. Illustration of how different the tip rate methods interpret speciation rate variability on a single simulated phylogenetic tree. DR approximates speciation rates as inverted values of species isolation progressively downweighted by distance from each focal tip, which was shown to converge to maximum likelihood estimate of speciation rate on pure birth trees ( Jetz et al. 2012 ). BAMM (Bayesian analysis of macroevolutionary mixtures; (Rabosky 2014) assumes that speciation rates rarely punctually change during evolutionary time, and the rest of the variation in the rates is gradual. ClaDS ( Maliet et al. 2019 ) assumes that the speciation rates change at every cladogenetic event but are still auto-correlated across the tree. Warmer colours denote faster estimated speciation rate. A problematic situation arises when these methods do not give compatible results. In such cases, the lack of external validation makes it impossible to untangle which of the assumed mechanisms of diversification variation are more appropriate, and determine which results are more credible. Even if the methods give similar estimates, which is fortunately often the case in empirical scenarios ( Rabosky 2014 ; Igea and Tanentzap 2020 ; Machac 2020 ), accuracy and statistical performance of these estimates is still unknown. Title and Rabosky (2019) and Malliet et al (2019) explored errors associated with BAMM, ClaDS and DR on phylogenies simulated using various forms of rate variability, such as gradually evolving rates, multiple rate regime shifts, and diversity dependence. However, it is uncertain how well these simulated phylogenies reflect empirical ones. The mechanisms they implemented in the simulations use ideas about rate variability common in the field, but are essentially the same ones that underlie the tested diversification methods. A test using the full complexity of empirical data is thus needed to evaluate the real-world performance of these methods and adequacy of the evolutionary mechanisms they assume. A potentially powerful empirical test of these diversification methods would explore whether their tip diversification estimates at one point in time accurately predict the diversity that individual tips generate subsequently. This is, of course, not possible in the present, as we do not have any information about the future. However, an approach adopted by Cantalapiedra et al. (2019) used phylogenies containing fossilized extinct species, calculated tip rates at arbitrary time slices and considered the “future” time as an arbitrary subsequent time bin. In their conservation-oriented study, they found that prioritizing species with the highest DR metric in the past led to slightly higher total diversity in subsequent time bins. This finding suggests that the DR metric has some predictive capacity, but the paper did not attempt to quantify this nor did it compare the predictive capacity of DR against other tip rate estimation methods. Another important limitation of this study was the exclusive reliance on fossil phylogenies. While they represent a data type that can be comfortably analyzed in a through-time manner, the fossil phylogenies are an incomplete representation of the past diversification process, and are known to be biased by nonrandom geographic distribution of fossil beds ( Flannery-Sutherland et al. 2022 ), different fossilization probabilities among taxa, and other related effects. In this paper, we perform cross-validation tests of commonly used tip diversification methods (BAMM, CLaDS and DR) on temporal slices of both fossil and extant phylogenies ( Fig 2 ). To do this, we developed an approach that mimics a generalized linear model, allowing us to link the tip estimates of diversification rates at a time slice with distributions of the number of subsequently surviving daughter species. This approach allows us to evaluate the relationship between estimated tip rates and realized rates of diversification in the subsequent slices of the tree, giving statistically meaningful and comparable measures of predictive capacity, such as generalized R 2 . Importantly, due to conditioning of the number of daughter species distribution, our approach can be used with both fossil and extant phylogenies. Both these data types have important advantages and constraints - the fossil phylogenies are an incomplete and nonrandom representation of the past diversification but carry explicit information about extinctions, and the extant phylogenies are more representative but contain only strongly censored information about the extinctions. Analyzing fossil and extant phylogenies in parallel allows us to exploit the advantages of each data type and draw conclusions that are less dependent on their limitations. Download figure Open in new tab Figure 2. Schematic representing the temporal cross validation methodology used in this study for fossil (left) and extant (center) phylogenetic trees. The phylogenetic trees were sliced at a certain point of time, representing “present” in the cross validation test. The part of phylogeny preceding the slice, representing “past”, was used to estimate tip rates of speciation and extinction for all branches at the slice using BAMM, CLaDS or DR. For each phylogeny branch at the slice, the number of subsequent daughter species after the prediction window was counted, representing the “near future”. Note that this daughter species count can be 0 for fossil phylogenies, but is >0 for extant phylogenies. Subsequently the daughter species counts were linked to the tip estimates of diversification using a likelihood based model (right, BirDLinG). This model assumed that numbers of daughter species were drawn from a distribution reflecting a homogeneous birth-death process with speciation and extinction parameters controlled by linear transforms of tip estimates of speciation and extinction rates for each tip at the slice. Materials & Methods The cross validation test linking tip rate estimates with subsequent diversity In our cross validation test, we used extant and fossil phylogenies and sliced them at particular points of time, dropping the lineages after this point ( Fig. 2 ). The slice time thus represented “present” in our test. On these truncated trees, representing the “past”, we employed the tip rate methods (BAMM, CLaDS and DR) to estimate diversification rates for each tip lineage at the slice time. For BAMM and CLaDS we estimated both speciation and extinction rates, and we considered DR to be an approximation of speciation rate ( Jetz et al. 2012 ) and its extinction rate estimates to be 0. To represent the “near future”, we counted the number of surviving daughter species from each tip after a prediction window of length t. The length of the prediction window was 5 My for fossil phylogenies, and 1, 5 and 10 My for extant phylogenies, where slices were positioned so that the end of prediction window was always at 0 My BP. Then we linked the tip rate estimates at slice point (“past”) and the subsequent number of daughter species (“near future”) using a custom developed likelihood-based approach called BirDLinG (standing for Birth-Death Linear Generalisation, https://github.com/smyckaj/BirDLinG ). BirDLinG assumes that the number of daughter species descending from a tip in a subsequent time bin of length t can be locally approximated by a time-homogeneous birth-death process with speciation rate λ and extinction rate μ. The probability distribution of number of surviving tips n resulting from a time-homogeneous birth death process after a prediction window of length t is: for n=0, and as described in Raup (1985). This equation can be used to determine the likelihood that the observed number of daughter species descending from a tip after time t was produced by any given values of λ subs and μ subs , reflecting the actual speciation and extinction rates subsequent to the slice. The total likelihood across all phylogeny tips of one slice can be consequently expressed as a product of such individual tip likelihoods. In BirDLinG, we assume that the vectors of subsequent speciation and extinction rates for all tips of one slice λ subs and μ subs are linear transformations of speciation and extinction rates estimated by tip rate methods λ tip and μ tip , so that And where the values of α λ , α μ , β λ , β μ are obtained using numerical likelihood optimization. With this formulation, we thus assume that there is a linear relationship between the tip estimates of diversification at the slice and the actual diversification rates generating the numbers of daughter species subsequent to the slice, and the direction of this relationship is controlled by slope parameters (β). Importantly, this formulation also accounts for the possibility that some of the tip rate methods systematically misestimate the subsequent diversification rates by a constant (α) or a proportion (β) but remain a linear transform of these values, which is still sufficient for many macroevolutionary applications. Also, formulated this way, our model mimics the structure of a generalized linear model (GLM), where the response variable n is drawn from the distribution defined by Equation 1, while the optimal parameters of this distribution λ subs and μ subs are latent and obtained via linear regression from a set of predictors λ tip and μ tip . For the slices of extant phylogenies, our model formulation followed the same logic as above, but we conditioned the probability distribution in Equation 1 to contain at least one species (n>0), accounting for the fact that branches with no surviving offspring lineages would not have been observed as a tip in the sliced tree. Importantly, this conditioning only works properly if the end of our prediction window (slice+t) is a 0 My BP from which the extant molecular trees were constructed ( Fig. 2 ). Putting it earlier in time while using the extant phylogenies would erroneously omit lineages that were surviving at the end of the prediction window but went extinct later towards the present. This limitation prevented us from slicing the extant phylogenies in a similar through-time manner as the fossil trees, so we focus here only on slices in the last 1, 5 and 10 My BP. Another limitation of using the extant phylogenies lies in the fact that censoring 0 values from species counts effectively removes the majority of information about extinctions from our response variable. We still use the extinction component of BirDLing for exploring the extant phylogenies rather than forcing α μ and β μ to 0. We do it because the predictor tip extinction rates might still be, in theory, meaningful, and future species counts depend on net diversification rates rather than only speciation rates. However, we are aware that the estimates of these parameters are not interpretable on their own. It is important to point out that in BirDLinG, we ignore the heterogeneity in the birth-death process among the daughter lineages in the prediction window subsequent to the slice, which is a relatively strong and potentially problematic assumption. Although this assumption follows the intuitive interpretation of tip rates in conservation biology ( Jetz et al. 2012 ; Cantalapiedra et al. 2019 ), it, technically speaking, is at odds with the assumed rate heterogeneity prior to the slice. For the parametric methods (ClaDS and BAMM), it would be theoretically possible to implement the prediction window cladogenesis to follow the pre-slice processes, using frameworks like Generative Bayesian Models ( Senderov et al. 2023 ). However, this approach would prevent us from testing DR, which, unlike CLaDS or BAMM, does not mechanistically address the pre-slice cladogenesis. We address this issue by two strategies: First, we only slice the phylogenies so that the pre-slice part of the tree is at least 3 times longer than the prediction window, allowing us to conclude that the prediction window is short enough to be approximated by a constant process. Second, we replicate our BirDLinG results with a Spearman rank correlation coefficient. This approach delivers very similar results (Supp Fig. Spearman), and the squared Spearman’s Rho can be interpreted in a similar way as the generalized R 2 values developed below. The Spearman correlation omits specific assumptions of BirDLinG, such as time homogeneity of the birth-death process in the prediction window and the linear relationship between tip estimates and true diversification rates. However, this generalisation comes at the expense of a parametric understanding of the system and also the potential to integrate tip estimates of speciation and extinction into a single predictive measure. Predictive accuracy measures We used the likelihood values of the BirDLinG models to calculate generalized R 2 , a measure commonly used in generalized linear models as an equivalent to Pearson’s coefficient of determination in normal linear models ( Cohen et al. 2015 ; García-Portugués). To calculate the generalized R 2 , it is necessary to compare the full model likelihood (L alt ) with a null likelihood (L null ) and a saturated likelihood (L sat ). The null likelihood can be obtained using the same reasoning as the full model likelihood described above, but with β λ and β μ equal to 0, i.e. with subsequent diversification rates λ subs and μ subs independent of tip estimates λ tip and μ tip . The saturated likelihood is the best obtainable likelihood for the given number of species. The saturated likelihood for individual tips is the maximum of the function from Equation 1. For n =0 it is at λ =0 and μ →∞ and equals to 1. For n =1, it is at λ =0 and μ =0 and also equals to 1. For n >1, it is at λ =log ( n )/ t and μ =0, and equals the value of Equation 1 for respective λ and μ (Raup 1985). The total saturated likelihood is a product of the individual tip likelihoods. The generalized R 2 using alternative, null and saturated likelihood is calculated as follows where the numerator and denominator of the fraction are deviances of alternative and null models, respectively. Our formulation of R 2 allows us to tackle the information flow from tip rate predictions towards the number of surviving lineages in a statistically meaningful and comparable way; however it cannot be interpreted as fully equivalent to the R 2 of a normal linear model. Notably, even if the tip estimates of diversification were perfect predictors of subsequent rates, it would not lead to R 2 =1 ( Hensher and Stopher 1979 ), because the variance of the birth-death distribution depends non-trivially on both these parameters. To ease the interpretation of empirically obtained R 2 , we thus calculate the maximum achievable R 2 by simulating n using a birth-death process with empirical λs and μs ( Supp. Fig. 1 ). Download figure Open in new tab Supplementary Figure 1. Theoretical R 2 generated by the three methods. Colors represent the methods (yellow = BAMM, red = ClaDS, blue = DR). Rows depict different taxon in the order: Ruminantia, Carnivora, Dinosauria from the top. The x-axis represents the time from the present and relates to the time slices from the respective phylogenies. The boxplots represent medians, quartiles and interquartile ranges across 100 simulated sets of species counts. Tip estimates of diversification rates The tip rate estimation methods compared in this manuscript are BAMM (Rabosky 2014), CLaDS ( Maliet et al. 2019 ) and DR metric ( Jetz et al. 2012 ). The BAMM model was fitted using the standard Metropolis coupled Markov chain Monte Carlo algorithm implemented in this method. The priors for each tree and time slice were generated with the function setBAMMpriors in the package BAMMtools (Rabosky 2014). Models were run with four chains at 100,000,000 iterations with 10% burn-in. Convergence was assessed with effective sample size (ESS >200) of the number of shifts. For the CLaDS model, we used its implementation in PANDA package in Julia ( https://github.com/hmorlon/PANDA.jl ), because it is considerably faster than the original R implementation. Models were run with three MCMC chains with 25% burn-in and otherwise default settings. Convergence was assessed after every 1000 iterations with the Gelman statistic ( Gelman et al. 2014 ), and the run was stopped once all the parameters had Gelman statistics < 1.05. DR was calculated as an inverted value of equal split measure ( Redding and Mooers 2006 ) estimated using R package picante ( Kembel et al. 2010 ). The DR metric is proven to be an unbiased approximation of speciation rate in a pure birth process ( Jetz et al. 2012 ), so we used it as a tip estimate of speciation and considered tip estimates of extinction from DR to be 0. This effectively leads to cancelling parameter β μ from Equation 2. The BirDLinG models for DR were thus fitted with only constant extinction rate α μ , and no variation in extinction among the tips. Phylogenetic datasets To test the predictive capacities of tip diversification metrics we used three phylogenetic datasets of tetrapod taxa including fossil tips. The fossil phylogenies of Ruminantia and Dinosauria are taken from Cantalapiedra et al. (2019) , and the phylogeny of all known Carnivora comes from Faurby et al. (2024) . We limited our inference to these three fossil datasets because all the other available fossil phylogenies of tetrapods (e.g., in Cantalapiedra et al. 2019 or Quintero et al. 2024) are too small for efficient tip rate inference (e.g., Equidae or Cetacea) and/or a taxonomic subset of the dataset we use (Caninae within Carnivora). All trees are resolved at the species level and are significantly informed by fossil data where phylogenetic relations of fossil taxa are well established. All datasets contain both extant and extinct species except the Dinosauria which spans a period of 181.2 to 66 million years, after which the lineage leading to modern birds is excluded. 10 trees were randomly sampled from a posterior sample of 100 trees for each dataset and used for downstream analyses. Trees that include dated fossil tips have the advantage that they contain explicit information on the lineages that went extinct, rather than just recording diversification surviving to the present. However, at the same time, the lineages contained in the fossil phylogenies are known to suffer from sampling biases related to nonrandom geographic distribution of known fossil beds ( Flannery-Sutherland et al. 2022 ) and different fossilization rates of lineages with different characteristics. We also used extant molecular phylogenies of birds ( Jetz et al. 2012 ), amphibians ( Jetz and Pyron 2018 ), mammals ( Upham et al. 2019 ), squamates ( Tonini et al. 2016 ), and turtles and crocodiles ( Colston et al. 2020 ). We chose these phylogenies because they are all relatively large (turtles and crocodiles being the smallest, having molecular information on 366 species and birds being the largest, with molecular information on 6670 species) and they all have sampling fractions above 50% (turtles and crocodiles having the highest at 95%). Like the fossil phylogenies, we randomly sampled 10 trees from the published posteriors to account for phylogenetic uncertainty. The fossil phylogenies were sliced every 10 million years and predictions were made 5 million years forward from every slice. This allowed us to identify possible differences in prediction ability due to the quality of more recent or ancient time scales while maintaining a consistent prediction window. Extant phylogenies were sliced at 1, 5, and 10 million years before present and (because these trees do not include information on extinction, see above) predictions were made to the present. Results The relationship between tip and subsequent estimates of speciation and extinction In order to understand the relationship between the tip estimates of diversification and the subsequent (latent) diversification rates, we inspected the slopes and intercepts of the linear components of our BirDLinG model. For the speciation rates, the relationships between subsequent rates and tip estimates are generally positive, with some exceptions, for BAMM, ClaDS and also DR ( Supp. Fig. 2 ). This shows that all three methods can in principle serve as better than random predictors of speciation rates at the tips. Importantly, similar patterns are apparent using Spearman correlations between tip rate estimates and subsequent number of species ( Supp. Fig. 3 ), suggesting that this general result is independent of specific assumptions in BirDLinG. Download figure Open in new tab Supplementary Figure 2. Histograms of the slopes generated by the three methods for fossil phylogenies. Columns represent the methods (left = BAMM, middle = ClaDS, right = DR). The top row depicts the slopes of all fossil phylogenies together while subsequent rows show the slopes of individual phylogenies in the order: Ruminantia, Carnivora, Dinosauria. Download figure Open in new tab Supplementary Figure 3. Spearman correlations generated by the three methods. Colors represent the methods (yellow = BAMM, red = ClaDS, blue = DR). Rows depict different taxon in the order: Ruminantia, Carnivora, Dinosauria from the top. The x-axis represents the time from the present and relates to the time slices from the respective phylogenies. The boxplots represent medians, quartiles and interquartile ranges across 10 samples from phylogenetic tree posteriors. A more detailed look at the magnitude and variation among the individual time slices and taxa, however, reveals that the relationship between tip estimates and subsequent rates of speciation is fairly context dependent ( Fig. 3 ): The speciation slope estimates for Dinosauria are consistently positive and aggregated around values close to 1 for BAMM and ClaDS, This shows that both methods retrieve correctly scaled estimates of speciation rates at the tips. The DR estimates for Dinosauria are also consistently positive, but are lower than 1, showing that DR tends to overestimate speciation rates. The speciation slope estimates for Carnivora and Ruminantia using BAMM and ClaDS are more often negative and also more variable, showing that neither method works optimally for these phylogenies ( Fig. 3 ). The most extreme examples are the slices of Ruminantia at 10 and 20 Ma BP, where the slopes are even consistently negative, i.e. the tips with highest speciation rate predicted by BAMM and ClaDS have the lowest subsequent speciation rates. In contrast to the two parametric methods, DR performs better on Carnivoran and Ruminant phylogenies, showing consistently positive slopes. However, these DR slopes are lower than in Dinosaurs, suggesting even more severe systematic overestimation. The speciation slope estimates in the extant phylogenies of Testudinae, Crocodilia and Amphibia using BAMM and ClaDS are generally positive and aggregated around 1, showing unbiased and correctly scaled estimation ( Fig. 4 ). The estimates using DR are positive but lower than 1, revealing that DR tends to overestimate the speciation rates also in the extant phylogenies. ADD DETAILS HERE WHEN WE HAVE COMPLETE RESULTS FOR MAMMALS, BIRDS AND SQUAMATES Download figure Open in new tab Figure 3. The links between tip estimates and true values of speciation (slopes) and the predictive capacities of different tip estimates (R 2 ), for BAMM (yellow), CLaDS (red), and DR (blue) at each slice in the fossil phylogenies. Phylogenies were sliced every 10 million years and predictions were calculated to predict the next 5 million years of diversification from the slice in the fossil-informed trees. The boxplots represent medians, quartiles and interquartile ranges across 10 samples from phylogenetic tree posteriors. Download figure Open in new tab Figure 4. The links between tip estimates and subsequent rates of speciation (slopes) and the predictive capacities of different tip estimates (R 2 ), for BAMM (yellow), ClaDS (red), and DR (blue) at each slice in the extant phylogenies. Phylogenies were sliced every 1, 5, and 10 million years and predictions were calculated to the present (i.e., the prediction windows are corresponding to 1, 5, and 10 million years). The boxplots represent medians, quartiles and interquartile ranges across 10 samples from phylogenetic tree posteriors.FINISH BIRDS, MAMMALS AND SQUAMATES The tip estimates of extinction, in contrast to speciation, do not appear to be meaningfully associated with subsequent extinction rates. This is universally true for all the examined fossil phylogenies ( Supp. Fig. 4 ), suggesting that neither BAMM nor ClaDS have capacity to correctly estimate extinction rate using phylogenetic trees. This is not surprising for ClaDS, where the extinction rates are resulting from speciation estimates and a global extinction fraction parameter - this implementation is clearly meant only to control for presence of extinction in the system, not for estimating extinction rates. On the other hand, the tip extinction rates in BAMM are implemented using an independent process of rate evolution similar to one governing speciation, and could in theory serve for forecasting, although it is discouraged by the authors of the algorithm ( Title and Rabosky 2019 ). Download figure Open in new tab Supplementary Figure 4. The links between tip estimates and subsequent rates of extinction (slopes). Colors represent the methods (yellow = BAMM, red = ClaDS, blue = DR). Rows depict different taxon in the order: Ruminantia, Carnivora, Dinosauria from the top. The x-axis represents the time from the present and relates to the time slices from the respective phylogenies. The boxplots represent medians, quartiles and interquartile ranges across 10 samples from phylogenetic tree posteriors. Predictive capacity of tip rate estimates We calculated generalized R 2 based on our BirDLinG models as a measure of the tip rate estimates capacity to predict subsequent diversification. For all BAMM, ClaDS and DR, the R 2 was typically between 0 and 30 % for both fossil ( Fig. 3 ) and extant ( Fig. 4 ) phylogenies, and ClaDS and DR exhibit slightly better R 2 values than BAMM ( Supp. Fig. 5 ). The amount of variation explained for Dinosauria, Testudinae, Crocodilia and Amphibia was up to 30%, and in phylogenies of Carnivora and Ruminantia up to 10% but typically lower. For the extant phylogenies of Testudinae, Crocodilia and Amphibia, there was no clear pattern among slices at 1, 5 and 10 My BP. Download figure Open in new tab Supplementary Figure 5. Histograms of the R 2 generated by the three methods for fossil phylogenies. Columns represent the methods (left = BAMM, middle = ClaDS, right = DR). The top row depicts the slopes of all phylogenies together while subsequent rows show the R 2 of individual phylogenies in the order: Ruminantia, Carnivora, Dinosauria. In order to map the maximum achievable values of R 2 in our BirDLinG model, we ran a series of simulations based on our empirical tip estimates of diversification. The maximum achievable R 2 for ClaDS and BAMM estimates range between 0% and 50%, and when compared to the empirical R 2 of individual phylogenies and slices (Supp. Fig. theoretical R2), they are typically larger by a factor of 1 to 4. This suggests that the tip rate estimates typically explain between 25% and 100% of the variation possible to explain, given the stochasticity inherent in the diversification process. The situation is more complicated in case of DR, because DR does not provide tip estimates of extinction. The shown maximum achievable R 2 ( Supp. Fig. 1 ) were simulated with μ=0, and are larger than the ones for BAMM and ClaDS. However, as μ is positively linked with variation in the birth-death distribution, these values can get arbitrarily lower by using larger μ values for the simulations. Discussion We have explored here the capacity of recently developed approaches for estimating tip rates of diversification (BAMM, ClaDS, and DR) to predict future diversification and diversity patterns. On average, all three methods produce tip estimates that are positively associated with subsequent speciation rates. This confirms that these methods are better predictors of future diversification than random guessing, a result that was previously demonstrated for DR on a more limited dataset of fossil-only phylogenies ( Cantalapiedra et al. 2019 ). On the other hand, our results also suggest that the amount of variability in future diversity patterns that can be explained by tip rate estimates is fundamentally limited. The R 2 values of empirical phylogenies were lower than 0.3, and the simulations suggest that a non-negligible part of variation in subsequent diversity patterns is not deterministically predictable due to the large amount of stochasticity in the speciation process and overall unidentifiability of the extinction rates. More worrying, although all the tip rate methods work satisfactorily on average, we found that certain combinations of methods and phylogenies produce estimates that are dissociated from subsequent dynamics, or even misleading. For instance in the phylogenies of Carnivora and Ruminantia, the BAMM and ClaDS tip estimates tend to be almost independent of subsequent diversification dynamics, and in two slices of Ruminantia they are even negatively associated with subsequent diversification. We provide two non-exclusive explanations of what could cause this local underperformance: First, it is possible that BAMM and ClaDS fail on Carnivorans and Ruminants simply because these phylogenies are smaller than the rest of the dataset. Parametric methods like BAMM and ClaDS require relatively large phylogenies to converge ( Maliet et al. 2019 ; Title and Rabosky 2019 ). While all slices of Carnivorans and Ruminants are larger than 30 species, which is commonly considered an adequate size for BAMM or ClaDS, they are still among the smaller phylogenies that we use in our study. Another possible explanation is that the Carnivoran and Ruminant phylogenies exhibit mechanisms of diversification rate inheritance that cannot be captured under the assumptions of ClaDS and BAMM. Both these methods expect that changes in the diversification rates are locally positively correlated along the tree. Perhaps there is a true lack of diversification inertia in the Carnivora and the inversion in Ruminantia is caused by some kind of saturation dynamics where the past radiation causes lack of speciation in the future (as described e.g. in Smyčka et al. 2023 ). These effects can be further amplified by potential misspecification of extinction rate dynamics in both BAMM and ClaDS. In the other phylogenetic datasets, that is Dinosauria, Testudines, Crocodilia and Amphibia, all BAMM, ClaDS, and DR give tip estimates of speciation that are positively associated with subsequent speciation rates. This suggests that an important component of diversification dynamics in these groups resembles the locally autocorrelated processes. Previous simulation results have shown that all three methods work well when the diversification rates shift according to a continuous Brownian motion, discrete rate-shift model, and several other locally autocorrelated mechanisms ( Maliet et al. 2019 ; Title and Rabosky 2019 ). Interestingly, ClaDS and DR mildly outperform BAMM in prediction capacity in Dinosaurs and the extant phylogenies. This might suggest that the modes of evolution with continuous or multiple small rate shifts as assumed by these methods are more represented in the empirical phylogenies than rarer large rate shifts modelled by BAMM, but it might similarly reflect known problems with BAMM model implementation ( Moore et al. 2016 ) or other issues. Another major difference among the methods is that DR shows a positive relationship with subsequent diversification more robustly than both ClaDS and BAMM. However, at the same time, DR also systematically overestimates tip speciation rates, which is in accordance with previous findings via simulations ( Title and Rabosky 2019 ). A possible interpretation of this phenomenon is related to the fact that both BAMM and ClaDS actively deal with extinction rates, whereas DR neglects extinction near the present ( Jetz et al. 2012 ), leading to a possible “pull of the recent” phenomenon. Including extinction into models of diversification thus might be necessary for retrieving correctly scaled speciation estimates, but comes at the expense of lower stability of the results. The tip estimates of extinction rates seem generally unrelated to the expected values of subsequent extinction rates reconstructed from fossil phylogenies. This is not surprising - reconstructing extinction rates at the tips of molecular phylogenies is generally considered an extremely complicated to impossible task ( Morlon 2014 ; Rabosky 2014; Louca and Pennell 2020 ). Moreover, the response variable of our approach - the species counts - carries a limited explicit information about subsequent extinction rates, as they are only indicated by the frequency of zeros in the fossil phylogenies. On the other hand, the generally poor performance of the extinction estimation is the likely reason why we see no major qualitative differences between speciation estimation and predictive performances on fossil and extant phylogenies - the whole system carries so little information about extinction that formally conditioning the response variable on non-extinction does not cause any additional harm. Importantly, the comparable results of the analyses of fossil and extant phylogenies suggest that our main conclusions are robust to the potential major sources of criticism of both data types, such as incomplete lineage sampling and general quality of the fossil data ( Flannery-Sutherland et al. 2022 ) and systematic omission of the lineages extinct after slicing in the extant phylogenies. This comparison shows that, when treated adequately, the temporal slicing experiments can be to some extent performed on the molecular phylogenies without extinct species. Our results have several interesting consequences for applications involving forecasting future diversification and diversity. We show that the tip rate estimates can in principle inform us about future speciation potential ( sensu Kling et al. 2018 ), but the prediction power of these estimates is severely limited. It is also important to point out that they only inform us about background evolutionary rates, uninfluenced by human impacts ( sensu Ceballos et al. 2015 ). On the practical side, we demonstrate that different methods of tip estimation have different properties and are potentially useful for different applications. In particular, DR is relatively robustly positively associated with subsequent speciation rate but systematically overestimates its value. Such robust relative measure is well tailored for tasks where preserving order is more important than the absolute values, such as giving species conservation priorities based on future potential ( Kling et al. 2018 ; Cantalapiedra et al. 2019 ). On the other hand, CLaDS and BAMM estimates are more unstable but centered around subsequent speciation rates. Estimates from these methods might be useful for comparing the current speciation rates based on phylogenetic trees with independent sources of evidence, such as the ones based on paleontological record ( Upham et al. 2021 ; Storch et al. 2022 ), microevolutionary dynamics ( Rolland et al. 2023 ) or biogeographic processes ( Faurby et al. 2022 ; Weil et al. 2025 ). An important problem with practical applications of tip rate metrics is that although they work on average better than chance, their statistical performance is widely variable based on individual phylogenies. In conservation applications, we may face a task where we have an extant phylogeny and want to predict future cladogenetic scenarios of individual species. Here, we would have no direct way to know whether we work with a low inertia phylogeny, such as Ruminants, allowing us to predict 1% of future diversification variability, or a high inertia phylogeny such as Dinosaurs allowing us to predict 10%. Luckily, our results show that the predictability is at least positively correlated among different slices of one phylogeny. In this scenario, we thus recommend using BirDLinG on a few slices of the past of an extant phylogeny to gain a rough idea of its predictability in the future. Predicting the future is always a complicated task, and predicting the future of a long term and extremely stochastic process, as is macroevolution, is extremely challenging. In this paper, we map the limits of tip rate methods, associated with specific methodological assumptions, to predict future diversities and diversification dynamics. Although complicated to do, we believe that the focus on predictive capacity of macroevolutionary hypotheses and mechanisms is justified. After all the predictions, both quantitative (e.g. species A will give rise to n daughter species) and qualitative (e.g., species are less likely to speciate when standing diversity is high), are one of the most practical deliverables that macroevolution may provide to the society and other scientific fields. We hope that this paper will help to spark more interest in the predictability of macroevolution and measuring the predictive capacity of hypotheses and mechanisms used in this field. Author contributions GIR and JS conceived the study. JS designed the BirDLinG with assistance from SPO, AOM and DS, and GIR performed the evolutionary analyses. GIR and JS drafted the first version of the manuscript, and DS, AOM and SPO further edited it. All authors contributed to the final version of the manuscript and GIR and JS contributed equally. Acknowledgements We thank A. Toszogyová, I. Šímová, D. Pelletier, S. Lambert and L. M’Gonigle for helpful comments and insights at different stages of the project. The research was funded by the Czech Science Foundation project GAČR 24-12851O to JS. GIR was also supported by GAČR 20-29554X to DS). Funder Information Declared Czech Science Foundation, https://ror.org/01pv73b02 , GAČR 24-12851O , GAČR 20-29554X Footnotes https://github.com/smyckaj/BirDLinG References ↵ Cantalapiedra J.L. , Aze T. , Cadotte M.W. , Dalla Riva G.V. , Huang D. , Mazel F. , Pennell M.W. , Ríos M. , Mooers A.Ø. 2019 . Conserving evolutionary history does not result in greater diversity over geological time scales . Proc. R. Soc. B Biol. Sci . 286 : 20182896 . OpenUrl CrossRef PubMed ↵ Ceballos G. , Ehrlich P.R. , Barnosky A.D. , García A. , Pringle R.M. , Palmer T.M. 2015 . Accelerated modern human–induced species losses: Entering the sixth mass extinction . Sci. Adv . 1 : e1400253 . OpenUrl FREE Full Text ↵ Cohen J. , Aiken L.S. , Cohen P. , West S.G. 2015 . Applied multiple regression/correlation analysis for the behavioral sciences . New York London : Routledge . ↵ Colston T.J. , Kulkarni P. , Jetz W. , Pyron R.A. 2020 . Phylogenetic and spatial distribution of evolutionary diversification, isolation, and threat in turtles and crocodilians (non-avian archosauromorphs) . BMC Evol. Biol . 20 : 81 . OpenUrl CrossRef PubMed ↵ Faurby S. , Pedersen R.Ø. , Svenning J.-C. , Antonelli A. 2022 . The counteracting effects of anthropogenic speciation and extinction on mammal species richness and phylogenetic diversity . Glob. Ecol. Biogeogr . 31 : 1810 – 1823 . OpenUrl ↵ Faurby S. , Silvestro D. , Werdelin L. , Antonelli A. 2024 . Reliable biogeography requires fossils: insights from a new species-level phylogeny of extinct and living carnivores . Proc. R. Soc. B Biol. Sci . 291 : 20240473 . OpenUrl CrossRef PubMed ↵ Flannery-Sutherland J.T. , Silvestro D. , Benton M.J. 2022 . Global diversity dynamics in the fossil record are regionally heterogeneous . Nat. Commun . 13 : 2751 . OpenUrl CrossRef PubMed García-Portugués E. Notes for Predictive Modeling . Version 5.12.1. ISBN 978-84-09-29679-8 . Available at https://bookdown.org/egarpor/PM-UC3M/ . ↵ Gelman A. , Hwang J. , Vehtari A. 2014 . Understanding predictive information criteria for Bayesian models . Stat. Comput . 24 : 997 – 1016 . OpenUrl CrossRef ↵ Gould S.J. , Eldredge N. 1977 . Punctuated Equilibria: The Tempo and Mode of Evolution Reconsidered . Paleobiology . 3 : 115 – 151 . OpenUrl Abstract ↵ Helmstetter A. , Zenil‐Ferguson R. , Sauquet H. , Otto S. , Mendez M. , Vallejo-Marin M. , Schönenberger J. , Burgarella C. , Anderson B. , de Boer H. , Glémin S. , Käfer J. 2023 . Trait‐dependent diversification in angiosperms: Patterns, models and data . Ecol. Lett . 26 : /a-n/a . ↵ Hensher D.A. , Stopher P.R. 1979 . Behavioural Travel Modelling . Croom Helm . ↵ Igea J. , Tanentzap A.J. 2020 . Angiosperm speciation cools down in the tropics . Ecol. Lett . 23 : 692 – 700 . OpenUrl CrossRef PubMed ↵ Jablonski D. 2008 . Species Selection: Theory and Data . Annu. Rev. Ecol. Evol. Syst . 39 : 501 – 524 . OpenUrl CrossRef Web of Science ↵ Jetz W. , Pyron R.A. 2018 . The interplay of past diversification and evolutionary isolation with present imperilment across the amphibian tree of life . Nat. Ecol. Evol . 2 : 850 – 858 . OpenUrl PubMed ↵ Jetz W. , Thomas G.H. , Joy J.B. , Hartmann K. , Mooers A.O. 2012 . The global diversity of birds in space and time . Nature . 491 : 444 – 448 . OpenUrl CrossRef PubMed Web of Science ↵ Kembel S.W. , Cowan P.D. , Helmus M.R. , Cornwell W.K. , Morlon H. , Ackerly D.D. , Blomberg S.P. , Webb C.O. 2010 . Picante: R tools for integrating phylogenies and ecology . Bioinformatics . 26 : 1463 – 1464 . OpenUrl CrossRef PubMed Web of Science ↵ Kling M.M. , Mishler B.D. , Thornhill A.H. , Baldwin B.G. , Ackerly D.D. 2018 . Facets of phylodiversity: evolutionary diversification, divergence and survival as conservation targets . Philos. Trans. R. Soc. B Biol. Sci . 374 : 20170397 . OpenUrl CrossRef PubMed ↵ Louca S. , Pennell M.W. 2020 . Extant timetrees are consistent with a myriad of diversification histories . Nature . 580 : 502 – 505 . OpenUrl CrossRef PubMed ↵ Machac A. 2020 . The Dynamics of Bird Diversity in the New World . Syst. Biol . 69 : 1180 – 1199 . OpenUrl CrossRef PubMed ↵ Maliet O. , Hartig F. , Morlon H. 2019 . A model with many small shifts for estimating species-specific diversification rates . Nat. Ecol. Evol . 3 : 1086 – 1092 . OpenUrl PubMed ↵ Moore B.R. , Höhna S. , May M.R. , Rannala B. , Huelsenbeck J.P. 2016 . Critically evaluating the theory and performance of Bayesian analysis of macroevolutionary mixtures . Proc. Natl. Acad. Sci . 113 : 9569 – 9574 . OpenUrl Abstract / FREE Full Text ↵ Morlon H. 2014 . Phylogenetic approaches for studying diversification . Ecol. Lett . 17 : 508 – 525 . OpenUrl CrossRef PubMed ↵ Rabosky D.L. 2014a . Automatic Detection of Key Innovations, Rate Shifts, and Diversity-Dependence on Phylogenetic Trees . PLOS ONE . 9 : e89543 . OpenUrl CrossRef PubMed Rabosky D.L. 2014b . Automatic Detection of Key Innovations, Rate Shifts, and Diversity-Dependence on Phylogenetic Trees . PLOS ONE . 9 : e89543 . OpenUrl CrossRef PubMed ↵ Rabosky D.L. 2022 . Evolutionary time and species diversity in aquatic ecosystems worldwide . Biol. Rev . 97 : 2090 – 2105 . OpenUrl ↵ Redding D.W. , Mooers A.Ø. 2006 . Incorporating Evolutionary Measures into Conservation Prioritization . Conserv. Biol . 20 : 1670 – 1678 . OpenUrl CrossRef PubMed Web of Science ↵ Rolland J. , Henao-Diaz L.F. , Doebeli M. , Germain R. , Harmon L.J. , Knowles L.L. , Liow L.H. , Mank J.E. , Machac A. , Otto S.P. , Pennell M. , Salamin N. , Silvestro D. , Sugawara M. , Uyeda J. , Wagner C.E. , Schluter D. 2023 . Conceptual and empirical bridges between micro- and macroevolution . Nat. Ecol. Evol . 7 : 1181 – 1193 . OpenUrl PubMed ↵ Ronquist F. , Kudlicka J. , Senderov V. , Borgström J. , Lartillot N. , Lundén D. , Murray L. , Schön T.B. , Broman D. 2021 . Universal probabilistic programming offers a powerful approach to statistical phylogenetics . Commun. Biol . 4 : 244 . OpenUrl PubMed ↵ Rosenzweig M.L. 2001 . Loss of speciation rate will impoverish future diversity . Proc. Natl. Acad. Sci . 98 : 5404 – 5410 . OpenUrl Abstract / FREE Full Text ↵ Senderov V. , Kudlicka J. , Lundén D. , Palmkvist V. , Braga M.P. , Granqvist E. , Çaylak G. , Virgoulay T. , Broman D. , Ronquist F. 2023 . TreePPL: A Universal Probabilistic Programming Language for Phylogenetics . :2023.10.10.561673. ↵ Simpson G.G. 1944 . Tempo and Mode in Evolution . Columbia University Press . ↵ Smyčka J. , Toszogyova A. , Storch D. 2023 . The relationship between geographic range size and rates of species diversification . Nat. Commun . 14 : 5559 . OpenUrl PubMed ↵ Storch D. , Šímová I. , Smyčka J. , Bohdalková E. , Toszogyova A. , Okie J.G. 2022 . Biodiversity dynamics in the Anthropocene: how human activities change equilibria of species richness . Ecography . 2022 :ecog.05778. ↵ Title P.O. , Rabosky D.L. 2019 . Tip rates, phylogenies and diversification: What are we estimating, and how good are the estimates? Methods Ecol. Evol . 10 : 821 – 834 . OpenUrl CrossRef ↵ Tonini J.F.R. , Beard K.H. , Ferreira R.B. , Jetz W. , Pyron R.A. 2016 . Fully-sampled phylogenies of squamates reveal evolutionary patterns in threat status . Biol. Conserv . 204 : 23 – 31 . OpenUrl ↵ Upham N.S. , Esselstyn J.A. , Jetz W. 2019 . Inferring the mammal tree: Species-level sets of phylogenies for questions in ecology, evolution, and conservation . PLOS Biol . 17 : e3000494 . OpenUrl CrossRef PubMed ↵ Upham N.S. , Esselstyn J.A. , Jetz W. 2021 . Molecules and fossils tell distinct yet complementary stories of mammal diversification . Curr. Biol . 31 : 4195 - 4206 .e3. OpenUrl CrossRef PubMed ↵ Weil S.-S. , Habrant M. , Lavergne S. , Allen W.L. , Gallien L. 2025 . Can macroevolution inform contemporary invasion potential? Biol. Invasions . 27 : 199 . OpenUrl View the discussion thread. Back to top Previous Next Posted October 07, 2025. Download PDF Data/Code Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Tip rate estimates can predict future diversification, but are unreliable and context dependent Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Tip rate estimates can predict future diversification, but are unreliable and context dependent Grace I Ridder , Jan Smyčka , David Storch , Arne Ø Mooers , Sarah P Otto bioRxiv 2025.10.06.680809; doi: https://doi.org/10.1101/2025.10.06.680809 Share This Article: Copy Citation Tools Tip rate estimates can predict future diversification, but are unreliable and context dependent Grace I Ridder , Jan Smyčka , David Storch , Arne Ø Mooers , Sarah P Otto bioRxiv 2025.10.06.680809; doi: https://doi.org/10.1101/2025.10.06.680809 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Evolutionary Biology Subject Areas All Articles Animal Behavior and Cognition (7618) Biochemistry (17635) Bioengineering (13859) Bioinformatics (41846) Biophysics (21401) Cancer Biology (18534) Cell Biology (25423) Clinical Trials (138) Developmental Biology (13352) Ecology (19860) Epidemiology (2067) Evolutionary Biology (24286) Genetics (15582) Genomics (22463) Immunology (17700) Microbiology (40298) Molecular Biology (17141) Neuroscience (88429) Paleontology (666) Pathology (2825) Pharmacology and Toxicology (4813) Physiology (7633) Plant Biology (15107) Scientific Communication and Education (2042) Synthetic Biology (4284) Systems Biology (9808) Zoology (2267)