iiinova — where your digital self keeps evolving

After testing fewer than 30 strains, we identified and validated a candidate in parC that conferred cross-resistance in the presence of gyrB D429N. We also found another lineage-specific pathway, highlighting the complexity of this clinically relevant trait. /7

A manhattan plot identifying a variant in parC

@sofiablomqvist.bsky.social then used BASS-GWAS to probe the genetic basis of previously described and unexplained cross-resistance between Zoliflodacin and Gepotidacin. This experiment required introducing the gyrB D429N variant into different N. gonorrhoeae strains, making it costly. /6

By efficiently integrating information from previous batches with the distribution of genetic variants, we find that BASS-GWAS consistently recovers causal variants in multiple real-world resistance traits with just 10s of isolates tested — a substantial decrease compared to random sampling. /5

Power curves for ciprofloxacin resistance variants

Power curves for high-level tetracycline resistance variants, downsampled to 10% and 2.5% phenotypic frequency

To tackle this, we set out to improve the efficiency of bacterial GWAS. BASS-GWAS uses Bayesian Adaptive Experimental Design and sparse whole-genome models to select batches of maximally informative isolates for phenotyping and applies to any measurable trait. /4

Calculations and existing studies show that fixed sampling requires 100s to 1000s of strains to power bacterial GWAS. Experiments at this scale are prohibitive and slow for many traits. High-throughput phenotyping tools do not generalize across traits and organisms. /3

Bacterial GWAS are a powerful tool for discovering the genetics underlying many traits. Past bacterial GWAS mainly used large clinical strain collections; sequencing was the main bottleneck. With large, sequenced strain collections now commonplace, phenotyping is the main limitation. /2

120

Bayesian adaptive experimental design for efficient microbial genome-wide association studies

How do we scale bacterial GWAS to traits that require (costly) experimental measurements? In our preprint, we: (1) introduce the highly-efficient Bayesian adaptive sequential sampling GWAS (BASS-GWAS); (2) use BASS-GWAS to study the genetics of cross-resistance between two new antibiotics! /1

biorxiv.org

Bacterial genome-wide association studies (GWAS) offer a powerful approach to identify the genetic basis of a trait measured in a set of sequenced isolates. As the number of sequenced isolates has gro...

152

Reposted by David Helekal

bioRxiv Genetics @biorxiv-genetic.bsky.social · 01/09/2026

Bayesian adaptive experimental design for efficient microbial genome-wide association studies www.biorxiv.org/content/10.64898/20…

054

Reposted by David Helekal

Aki Vehtari @avehtari.bsky.social · 23/06/2026

New paper "To select or not to select: predictively consistent priors instead of model selection" with Anna Elisabeth Riha, Leevi Lindgren, @davidkohns.bsky.social, @paulbuerkner.com arxiv.org/abs/2606.22850 Model selection is not a substitute for building good models in the first place 1/

To select or not to select: predictively
consistent priors instead of model selection

Anna Elisabeth Riha, Leevi Lindgren, David Kohns, Paul-Christian Bürkner, Aki Vehtari

Bayesian modelling workflows often consider multiple candidate models of varying complexity. Model selection is commonly used to navigate potential trade-offs between model complexity and generalisability to new data. We study when model selection is unnecessary or can even be harmful for predictive performance in finite data regimes and find that the need for selecting simpler models can depend on prior choice. We formalise predictively consistent priors, which keep prior predictive implications stable as model complexity increases. Across examples and numerical experiments, including adding covariates in linear and logistic regression, forward variable selection, and nonlinear modelling, flexible models with predictively consistent priors typically match or outperform selected simpler models in out-of-sample predictive performance. When selection helps, it can indicate poor joint prior implications, such as excessive prior mass on implausible predictive values. Based on our findings, we propose replacing the notion of sparsity or parsimony at the level of model components with specifying priors that remain sensible in predictive space as models become more complex.

110631

My big hope is that someday we will get community and environmental sampling for ESKAPE pathogens, as I do not see any other way to understand what shapes AMR for those bugs.

030

Moreover, for many infections, cultures are performed only after treatment failure. The data quality and integrity were one of the big reasons for working on GC as a sort of a 'model' organism. Speaking of GC, I think higher-level azithromycin resistance there falls under the mechanism you propose.

010

When it comes to resistance data, we often have severely biased datasets. E.G., for ESKAPE pathogens the data usually comes from bloodstream infections, and may be enriched for resistance (or even worse, virulence and resistance). Therefore, we're missing most of the transmission cycle.

The distribution of antibiotic use and its association with antibiotic resistance - PubMed

On the usage data side, we usually have data aggregated too coarsely, and thus cannot properly assess exposures. Moreover, that data is often very unreliable. The distribution of antibiotic use is also quite 'heavy tailed' pubmed.ncbi.nlm.nih.gov/30560781/.

pubmed.ncbi.nlm.nih.gov

Antibiotic use is a primary driver of antibiotic resistance. However, antibiotic use can be distributed in different ways in a population, and the association between the distribution of use and antib...

I very much agree that the mechanism proposed by your model is a major force shaping resistance. I think it's very hard to determine what is shaping resistance (your model, vs directional selection, vs something else) given the exceptionally poor data we usually have.

In most cases, multiple antibiotics are in use, and this is always the case. We therefore usually also have multiple resistant strains/lineages competing, each resistant to different antibiotic classes.

000

In our work on N. gonorrhoeae, we documented that mutations conferring the same resistance phenotype can have wildly different fitness costs. We saw this both in a phlyodynamic estimate, and an in vitro competition experiment. (From doi.org/10.1038/s415...)

100

... as well as with different resistance patterns. For example resistance to penicillins can be acquired through a beta-lactamase or mutations in penicillin binding proteins. Resistance to fluoroquinolones can be acquired through different resistance mutations in one of several genes.

Sorry, to clarify i was thinking clonal interference reducing efficiency of selection, and this reduction further amplified by fluctuating pressures. I think clonal interference is quite common in AMR. Often there are multiple strains / lineages competing, both with the same resistance pattern ...

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

Reposted by David Helekal

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

David Helekal @dhelekal.bsky.social · 19/06/2026

Not to mention, that resistance to different drugs within one class can be hugely variable. E.g. there's a world of difference between carbapenems, third gen. cephalosporins, and second gen. cephalosporins, despite all of these being classed as non-penicillin beta-lactams.

100

David Helekal @dhelekal.bsky.social · 19/06/2026

E.g. fluoroquinolone or tetracyline resistant Neisseria Gonorrhoeae in south east asia. With fluctuations in use we would at the very least expect some version of clonal interference in a fluctuating environment on top of mut-sel balance.

200

David Helekal @dhelekal.bsky.social · 19/06/2026

How much do we actually believe that the antibiotic exposures are constant? Often, we are restricted to very poor proxies for antibiotic exposure, such as population use pooled by drug type, or even worse, hospital use. We also know that in use settings ~approximate fixation is very much possible

200

Reposted by David Helekal

John Lees @johnlees.bacpop.org · 16/06/2026

We've been looking at how to compare and cluster large numbers of genomes, such as those in large isolate databases such as AllTheBacteria, and metagenome assemblies (e.g. SPIRE, MGnify). On a combined dataset of 5.6 million assemblies, we can now cluster/dereplicate everything in under a day!

13111

Reposted by David Helekal

The Lancet Microbe @lancetmicrobe.bsky.social · 10/06/2026

New comment Genotype-based antimicrobial resistance tests need a new regulatory framework www.thelancet.com/journals/lan... #IDSky #ClinMicro #AMR #OpenAccess #OA

Principles of genotype-based AMR diagnostic regulation
1.
Contextualise performance claims: require stratified performance, where feasible, by geography, time window, and phylogenetic lineage, and transparent isolate sampling frames. Intended-use claims should match demonstrated coverage.
2.
Make representativeness an evidentiary criterion: treat representativeness like sample size—as a formal element of evidentiary sufficiency. For broad indications, requirement of minimum coverage across regions, lineages, or demographic groups. When gaps remain, either narrow labelling or use staged indication expansion tied to new data.
3.
Mandate prespecified post-market monitoring and re-evaluation: approval should be contingent on minimum confirmatory phenotyping rates among susceptible calls (potentially risk-stratified and geography-stratified), sentinel networks for intensive culture or phenotyping, and predefined triggers (eg, rising very major errors, detection of novel determinants) that prompt modification and review.
4.
Create pathways for iterative updates: because rules or models should evolve, regulators should provide streamlined mechanisms for updating genotype–phenotype rules and algorithms, with clear expectations for evidence and re-evaluation. These pathways for iterative updates could reduce disincentives to improve tests rapidly.
5.
Integrate with AMR surveillance systems: these diagnostics are at the intersection of clinical care and surveillance. Regulatory requirements should align with national and global AMR surveillance to support data sharing, standardised definitions, and rapid recognition of emerging mechanisms.

0119

Reposted by David Helekal

Vittoria Colizza @vcolizza.bsky.social · 19/04/2026

Great discussion with @billhanage.bsky.social and the audience at #ESCMIDGlobal2026 during our session “Pathogens and people: modelling epidemics for preparedness”. Strong focus on how to improve preparedness in collaboration with public health institutes - lots to take forward.

163

Reposted by David Helekal

arXiv q-bio.PE Populations and Evolution @qbiope-bot.bsky.social · 15/04/2026

Ian Braga, Lucas Wardil, Ricardo Martinez-Garcia: Fixation probabilities for multi-allele Moran dynamics with weak selection arxiv.org/abs/2604.12004 arxiv.org/pdf/2604.12004 arxiv.org/html/2604.12004

012

Reposted by David Helekal

CIDRAP @cidrap.bsky.social · 07/04/2026

The European Commission will invest $34.7 million in the two public-private partnerships to boost the pipeline for new antibiotics and other AMR medical countermeasures. Read more: ow.ly/Sap850YFgir

22810

Reposted by David Helekal

Ana Ignatieva @anaignatieva.bsky.social · 24/03/2026

We (with @yanwong.bsky.social) are advertising a 3 year postdoc position for a project on ARG inference and methods! Based at Oxford stats. Closing date noon 7 April, full details at my.corehr.com/pls/uoxrecru...

my.corehr.com

Job Details

078

Reposted by David Helekal

Matthew J Shepherd @matthewjshepherd.bsky.social · 20/02/2026

🚨 New pre-print! 🚨 In the largest study of its kind to-date, we investigate the ecological and evolutionary mechanisms driving within-patient evolution of antimicrobial resistance (AMR). Read here: www.biorxiv.org/content/10.6... , and follow along with this thread, discussing our findings (1/21)

biorxiv.org

26533

Reposted by David Helekal

Richard Everitt @bayesianstats.bsky.social · 09/02/2026

Ian Roberts published his first paper "Bayesian Inference of Pathogen Phylogeography using the Structured Coalescent Model" with me, Jere Koskela and Xavier Didelot. We describe an efficient reversible jump MCMC method for inference for the structured coalescent. journals.plos.org/pl...

1111

Reposted by David Helekal

Yonatan Grad @yhgrad.bsky.social · 30/01/2026

Now published: our work using phylodynamics from surveillance data to quantify and experimentally validate the fitness impact of antibiotic resistance determinants & how this changes with patterns of antibiotic use: www.nature.com/articles/s41...

nature.com

Quantifying the real-world impact of antibiotic use and genetic determinants of resistance on gonococcal dynamics

Nature Microbiology - Phylodynamic modelling shows how the changing antibiotic landscape and genetic determinants of resistance shape real-world gonococcal dynamics. Experiments validated that...

16531

Reposted by David Helekal

Michael Plank @michaelplanknz.bsky.social · 29/01/2026

Very happy to have had the opportunity to contribute to this work. Thanks to Robin Thompson for leading, the amazing team of co-authors and the INI for hosting us.

0158

David Helekal @dhelekal.bsky.social · 14/07/2025

✅ Say: "My manuscript has been metropolised" ❌ Don't say: "My manuscript has been rejected and so I've sent it to MDPI"