iiinova — where your digital self keeps evolving

Andrew Roger @andrewjroger.bsky.social · 21/09/2026

The brilliant Prof. Edward Susko gave a great online lecture on modeling heterogeneity in sequence evolution in the phylogenetic context as part of the @phyloseminar.bsky.social series: www.youtube.com/watch?v=-1aJ...

youtube.com

Phyloseminar #157: Ed Susko (Dalhousie University)

YouTube video by phyloseminar.org

074

Andrew Roger @andrewjroger.bsky.social · 11/09/2026

Great! I assume it is being updated now? Some really big discoveries in the past few decades have changed our understanding of the 'deep' ToL

001

Reposted by Andrew Roger

Sishuo Wang @sishuowang.bsky.social · 09/09/2026

[new preprint] By sampling new basal non-symbiotic members and applying mixture subs models (incl GTRmpix) in dating, we show a contemporaneous emergence of Bradyrhizobium nodulation and legumes using an approach that co-dates both symbionts and legumes. www.biorxiv.org/content/10.6...

biorxiv.org

Molecular clock dating under mixture substitution models indicates contemporaneous emergence of Bradyrhizobium nodulation and legumes anchored by non‑symbiotic relatives

The transition from free‑living bacteria to mutualistic symbionts is a major evolutionary innovation, but its timing and stepwise nature remain unresolved because non‑symbiotic ancestors have rarely b...

163

Reposted by Andrew Roger

Wayne Maddison @waynemaddison.bsky.social · 09/09/2026

The Tree of Life web project is back! A phylogenetic view of biodiversity way ahead of its time (ca. 1995!), conceived and led by David Maddison, collaborative & human-curated, widely used in teaching. Defunct technology had doomed it until Travis, Danny, and Andrew revived it. 🥳🎉 🧪 #evolbiol

010250

Andrew Roger @andrewjroger.bsky.social · 01/09/2026

Yeah, and Randolph Kirkpatrick was probably right too...all of these rocks everywhere are fossil relics of the original houses of quasi-lifelike things living in tiny holes in hydrothermal vents

010

Andrew Roger @andrewjroger.bsky.social · 01/09/2026

Another groundbreaking paper from these wizards.

031

Andrew Roger @andrewjroger.bsky.social · 28/08/2026

That's almost the same thing as my cats hating Rammstein. 😛

010

Andrew Roger @andrewjroger.bsky.social · 21/08/2026

No time to diverge OR From Russia with Likelihood

150

Andrew Roger @andrewjroger.bsky.social · 21/08/2026

Very true. I'll try them on some of your G&T tunes next!!

010

Andrew Roger @andrewjroger.bsky.social · 20/08/2026

Does anyone else have the problem that their cats don't like Rammstein?

450

Reposted by Andrew Roger

American Society for Cell Biology @ascbiology.bsky.social · 24/07/2026

From MBoC A new localization atlas maps more than 600 Giardia proteins, revealing the locations of hundreds of previously uncharacterized proteins and creating a valuable resource for functional discovery. www.molbiolcell.org/doi/10.1091/mbc…

22311

Reposted by Andrew Roger

Molecular Biology and Evolution @molbioevol.bsky.social · 03/08/2026

Záhonová et al. tested whether hydrogenosomes and mitosomes in Metamonada originated from an independent endosymbiosis, finding instead that these organelles originated from an Alphaproteobacterium during a single common endosymbiosis. 🔗 doi.org/10.1093/molbev/msag175 #societyjournal #compbio

doi.org

Phylogenetic ancestry of Metamonada proteins points to a common origin of mitochondria in all eukaryotes

Abstract. Hypotheses concerning eukaryogenesis, the evolution of eukaryotic cells, differ in the relative timing of mitochondrial acquisition. Recently, a

074

Reposted by Andrew Roger

Dr John Clarke @jclarkepaleo.bsky.social · 03/08/2026

Our new paper introduces a fast & Bayesian state-dependent OU model in RevBayes that implements joint inference, to test adaptive macroevolutionary hypotheses. Congrats @prilau.bsky.social! Be sure to give her a follow & give her original tweet some love! Many more exciting PCMs in her future!

0167

Reposted by Andrew Roger

Joanna Masel @joannamasel.bsky.social · 27/07/2026

Our paper on how better amino acid substitution matrices trained on cleaned alignments improve gene tree inference, and on best practice for cleaning MSAs more broadly, is now out academic.oup.com/mbe/advance-.... New options in both IQTree and Muscle5. @phylowheeler.bsky.social

academic.oup.com

Improved gene tree inference from removing alignment errors both from focal genes and when training substitution models

Abstract. Multiple Sequence Alignment (MSA) is a key step in phylogenetic analysis and is prone to error. Unfortunately, algorithms that remove likely alig

07839

Andrew Roger @andrewjroger.bsky.social · 25/07/2026

A really nice essay. Worth reflecting on.

050

Reposted by Andrew Roger

Bornberglab @bornberglab.bsky.social · 28/01/2026

New review out in Nat Rev Genet: Emergence & evolution of protein-coding de novo genes by Erich and Lars Eicholt @lacholt.bsky.social. How non-coding DNA becomes translated, persists or is lost in populations, and can yield structured/functional proteins—plus methods & open questions! rdcu.be/e09SM

rdcu.be

Emergence and evolution of protein-coding de novo genes

Nature Reviews Genetics - De novo gene evolution entails the birth of new genes from previously non-coding DNA. In this Review, Bornberg-Bauer and Eicholt overview how protein-coding de novo genes...

1208

Andrew Roger @andrewjroger.bsky.social · 19/06/2026

11/ Also, it's easy to find, but to make it easier, here is the IQ-TREE site to download v3.1.3 from: iqtree.github.io

iqtree.github.io

IQ-TREE: Efficient phylogenomic software by maximum likelihood

010

Andrew Roger @andrewjroger.bsky.social · 19/06/2026

9/ The good news is that the SXXpfamCYY series are including in the latest version of IQ-TREE3 (v3.1.3) so you can try them yourself! This work was led by the brilliant Ryo Harada @ryoharada.bsky.social along with many collaborators, including Minh Bui, Rob Lanfear and the IQ-TREE team.

120

Andrew Roger @andrewjroger.bsky.social · 19/06/2026

8/ GTRspmix is part of a broader push toward models that better reflect how proteins evolve on the billion-year time scale: site preferences, site rates, exchangeabilities, and variation in substitution dynamics over the tree needs to be modeled in a computationally efficient way.

100

Andrew Roger @andrewjroger.bsky.social · 19/06/2026

7/ We also estimated general-purpose empirical GTRspmix models from Pfam: the SXXpfamCYY series. These are designed to make the approach usable without re-estimating complex model components for every analysis. The best performing overall model was S28pfamC59.

100

Andrew Roger @andrewjroger.bsky.social · 19/06/2026

6/ We evaluated GTRspmix on empirical and simulated protein datasets. The goal was not just a better likelihood score, but better behaviour in the kinds of difficult phylogenetic problems where model realism matters most.

100

Andrew Roger @andrewjroger.bsky.social · 19/06/2026

5/ Why does this matter for deep evolution? Over very long timescales, weak model violations can accumulate into strong support for the wrong tree. Better models are one way to avoid such phylogenetic artifacts. GTRspmix models are better: they fit real data much better than profile mixture models.

130

Andrew Roger @andrewjroger.bsky.social · 19/06/2026

4/ GTRspmix relaxes that assumption. It combines multiple exchangeability matrices with profile and rate mixture models, so different classes of sites can have different substitution dynamics. It does this efficiently, avoiding excessive computational cost and overparameterization.

110

Andrew Roger @andrewjroger.bsky.social · 19/06/2026

3/ Site-profile mixture models (e.g. C10-C60 or CAT) already model different amino acid preferences at sites. Site-rate models (e.g. G4) capture different sites evolving at different speeds. But exchangeabilities—the pairwise substitution “rules” among amino acids—are usually shared across sites.

100

Andrew Roger @andrewjroger.bsky.social · 19/06/2026

2/ Most amino acid subsitution models assume the same exchangeability matrix across all sites: one set of relative rates of interchange between amino acid pairs. That is convenient, but real substitution dynamics are more complicated.

100

Andrew Roger @andrewjroger.bsky.social · 19/06/2026

1/ Deep-time phylogenetics is hard: overly simplistic substitution models can mislead tree estimation at the billion year timescale. Our new preprint introduces GTRspmix, a protein modeling framework designed to more realistically model site-to-site heterogeneity in amino acid replacement. 🧵

15221

Reposted by Andrew Roger

Rowan Green @rowancallumg.bsky.social · 16/06/2026

Where are mutations most likely to happen in the genome? @mattjago.bsky.social has answers! This paper is also a masterclass in simple visualizations of complicated ideas. I really enjoyed getting deep into these results as a middle author 🧬

2247

Reposted by Andrew Roger

ISEP @isepprotists.bsky.social · 13/06/2026

New #ISEPpapers! Diversity and evolution of DNA polymerase θ in eukaryotes and the origin of mitochondrion-localized DNA polymerase PolIA in euglenozoa www.sciencedirect.com/science/arti... #Protists #Microbes #Evolution #Mitochondria

a schematic evolutionary tree of microbes illustrating a hypothesis discussed in the linked paper

152

Reposted by Andrew Roger

Pawel Burkhardt @pawelburkhardt.bsky.social · 16/06/2026

How ancient are the building blocks of animal sensory systems? Key components of animal sensory systems evolved before animals. We find that choanoflagellates possess diverse & spatially segregated TRP channels, pointing to ancient origins of sensory specialization. doi.org/10.64898/202...

412158

Reposted by Andrew Roger

ISEP @isepprotists.bsky.social · 04/06/2026

New #ISEPpapers by @deemteam.bsky.social! Reversal to osmotrophy in eukaryotes www.nature.com/articles/s41... "Distant eukaryotic lineages convergently reverted from predation to osmotrophy through co-option of bacterial genes and their mobilization via eukaryote-to-eukaryote HGT" #Protists

a simplified evolutionary tree of eukaryotes

13413

Reposted by Andrew Roger

Heng Li @lh3lh3.bsky.social · 30/05/2026

Jeremy Wang developed rammap, a minimap2 rewrite in Rust. It achieves comparable or better performance than minimap2 and produces identical output to minimap2. During rewrite, Jeremy found two long-existing bugs in minimap2 which are fixed in v2.31. www.biorxiv.org/content/10.6...

biorxiv.org

310944

Andrew Roger @andrewjroger.bsky.social · 02/06/2026

The bottom line is that you should always test profile mixture models amongst the set of models you are testing and check in the log file that they were actually tested. I recommend always trying C60+G, since any deeptime data set will usually find that model to fit best (probably with ELM or EAL)

120

Andrew Roger @andrewjroger.bsky.social · 02/06/2026

I'm reposting this because I want to make a correction. In the thread below I mention using the -mset flag for testing site profile mixture models. But in fact you should use -madd flag. I'm sorry for this mistake.

2137

Reposted by Andrew Roger

Molecular Biology and Evolution @molbioevol.bsky.social · 29/05/2026

Pisani, @elpiratavell.bsky.social et al. dissipate doubts about the utility of models accounting for compositional heterogeneity across sites and identify CAT-GTR as one of the most flexible models in the phylogenomic arsenal. 🔗 doi.org/10.1093/molbev/msag090 #evobio #molbio #phylogenomics

doi.org

Phylogenomic mixture models outperform homogeneous and partitioned models

Abstract. Significant advances have been made in resolving the tree of life, but many nodes remain debated. The last two decades saw the emergence of mixtu

0115

Andrew Roger @andrewjroger.bsky.social · 28/05/2026

11/ Huge credit to Charley McCarthy (@cgpmcc.bsky.social), who led this work, and to Ed Susko who implemented the models. Software and related scripts are available from Ed Susko’s software page: www.mathstat.dal.ca/~tsusko/soft...

mathstat.dal.ca

020

Andrew Roger @andrewjroger.bsky.social · 28/05/2026

10/ Take-home message: compositional heterogeneity is not just a nuisance—it can change phylogenetic conclusions. GFmix gives us a practical way to model some of that complexity in protein phylogenomics.

111

Andrew Roger @andrewjroger.bsky.social · 28/05/2026

9/ We then apply the models to a real dataset with known compositional-heterogeneity artefacts. The most complex fully optimized GFmix version consistently supports the correct tree over the artefactual tree, with improved likelihoods.

100

Andrew Roger @andrewjroger.bsky.social · 28/05/2026

8/ On simulated site-and-branch-heterogeneous data, the improved models better estimate branch-specific compositions and branch lengths. That matters because branch lengths and compositions are often entangled in difficult deep-time datasets.

100

Andrew Roger @andrewjroger.bsky.social · 28/05/2026

7/ We also introduce methods to detect compositional heterogeneity directly from sequence data, helping users decide when these more complex models are likely to be needed.

100

Andrew Roger @andrewjroger.bsky.social · 28/05/2026

6/ The improved GFmix framework also allows user-defined forms of compositional heterogeneity, not just GARP/FYMINK, and supports full maximum-likelihood optimization of the relevant parameters.

100