Lovely morning swim in Highgate ponds. Not sure how low a water temperature I am going to go to!
Executive Director EMBL. I have an insatiable love of biology. Consultant to ONT and Cantata (Dovetail)
7,138 followers 554 following 1,581 posts
view profile on Bluesky
Lovely morning swim in Highgate ponds. Not sure how low a water temperature I am going to go to!
Ha! No. But gloves are off when there are indels / CNVs …
And … you might think 1bp is the immutable unit but insertions and deletions means … well … it’s all about allele definition:)
For me the dominance / recessive is a special class of non linear effect which is dependent on the definition of alleles at a locus. If one allele == 1bp … compound hets LOFs are two alleles with epistasis; if my definition of LOF of gene its dominance/recessive
Today we are delighted to welcome Professor Samra Turajlić (@turajliclab.bsky.social) as our new Institute Director! Her work - spanning basic, translational & cancer research - has advanced our understanding of cancer biology & treatment Read more here: www.cruk.manchester.ac.uk/news/profess...
Wow. September has really lent into "its the start of autumn numpties" in the UK - I am looking out at driving rain. All good for aquifers, gardens etc - not so sure about harvests - All very ... ummm... British.
I was interviewed for ArtificiallyEverAfter, an AI podcast from the UK, on the genome and its impact on and its use of AI - going from the basics (what is DNA) to the ethics (what are the dangers of having AI methods on genomics). linktr.ee/artificially...
👋 Time to say goodbye to another one of our PhD students, @smuellerdott.bsky.social. She was working on the integration of phosphoproteomics and transcriptomics data and will now join the Computational Oncology Team at Merck. We wish her all the best for this new chapter! 🚀 ✨
One for the reading pile and this scQTL x Disease in a MR framework feels a v powerful approach (more tissues / cell types please!) -
Just brilliant. So well deserved - Rory has had a long career across both clinical trials and then cohorts (UK BioBank) and in my view his leadership and selflessness is a role model for researchers.
Hmmm. Genuinely I have given that talk to quite a few teenagers and not selective. I guess my experience is different from yours but I will stick to my guns - many - most - teenagers in my experience can understand the gist of this.
Whatever works. Actually I have a talk aimed at 14-18 year olds on this topic with these themes that seems to work well (definitely gets them thinking and asking questions) so ... I think people, including young people, can be pretty nuanced.
Again, do read the thread. My sense though from this reply is that you don't particular want to understand the relationship between these things, so ... I will now be blocking you. Have a lovely day.
No; do read the thread if you want it explained.
(I know this is tough for people. Ethnicity / Race is a big part of identity for most people and our society, and DNA/genetics feels like it is some sort of scientific equivalent of identity, but it is really worth ... separating. Tough I know for many people).
So untangle Ethnicity/Race from Genetics - I think it is far better to keep these concepts well separated in people's heads (without, of course, denying that people with black skin colour have different skin colour genetics from white people)
No matter what their skin colour, they are all British - and play for the English football team and I certainly would not make the cut.
Put in Cole Palmer - most people would label him as White British but has Afro-Carribbean grandparents - and Bukayo Saka - Black British, parents from Nigeria - for any part of the genome it is genuinely a complete mess about which one of us matches in terms of the suits for any particular card.
So - when you look at two people - myself and say Kyle Walker (Black British/mixed race central defender, now at Burnley) - our skin colour is different meaning you can guess we've got different suits (alleles) in the 30 numbers which are about skin colour but we've got 299,970 other numbered cards
Just because 30 cards (... the 7, the 281, and the Jack etc - these are the 30 places which strongly influence skin colour) are coming in mainly one suit (call it... hearts) it doesn't mean you can predict the suits well of any of the other numbers in someone's genetic hand.
So I might have a 1 of hearts and 1 of clubs at the start of chromosome 1, and then for the next position, the "2", perhaps I got a 2 of diamonds from both parents, and then the 3 I got a diamond and a club ... etc for 300,000 (split into 23 chromosomes, with special rules on sex chromosomes)
Everyone gets delt (via their parents) one card of each number (1 to 300,000) from each parent - your "genetic hand" are these cards twice - one from dad, one from mum. For this thought experiment, let's imagine the cards come in 4 suits (though this is not how it works in real life)
There are (crudely - it is rather more complex than this) 300,000 different parts of the genome to be shuffled - like a deck of cards with 300,000 numbers/picture cards (1 to 300,000 if you like) and each card comes in 2 to (top end) 100 different suits
Because someone has black skin means there are about ~30 places on the genome where you could have a good guess of their genome (though skin colour is surprisingly complex genetics at these 30 places) but it doesn't mean you can easily predict other places in their genome
This mixing proportion means that the small part of the genome behind visible characterisitics - skin colour, eye shape - moves around, and these visible characterisitics, although genetically determined is ... a very small part of the genome.
There are two things going on in human populations - many people stay relatively close to where they are born, and many people move. Because of those movements we mix *all the time* - in pre-history (we can see this in DNA), in ancient history and now.
The next thing to realise is that humans are (a) a very young species (we're crazy young) and (b) some of us have big wanderlust - as individuals and as groups. Sometimes the travelling is also forced on people (the most horrific being the transatlantic slave trade).
Go far enough back (~7 generations) you will have ancestors which did not contribute DNA to you (but might have to your sister or brother or cousin).
Each generation back you double the number of ancestors, 2, 4, 8, 16, 32 etc - meaning very quickly you have *alot* of ancestors. Because of the halving not every bit of DNA in the ancestors makes it the next generation.
For genetics every generation shuffles and halves the genome information they pass on (our chromosomes - how are DNA is stored - come in pairs; in our eggs or sperm we shuffle the genome between the pair on each chromosome and hand over half to our children).
Many visible characteristics (skin colour, eye shape etc) of humans is largely genetic. Some - like clothing, makeup, jewelery is very much not genetic. Just to say, despite the confident self assignment of people into race/ethnicity groups, it is actually impossible to have hard criteria for this.
Genetics is all of your DNA inherieted from your parents, and their parents (your grandparents) etc until we get to the population in (most likely) East Africa where H. sapiens evolved as the most successful in a series of bipedal apes.
Race (US term) or Ethnicity (UK term) is a way people self label (or sometimes others label) often anchored on visible characteristics, notably skin colour. You are presented with a form with a number of defined boxes and asked to tick one, with often the possibility of "Other"
An evergreen thread: Race/Ethnicity is *not the same* as genetics, and you can't use Race/Ethnicity as a sort of stand-in for genetics. These two concepts are connected via aspects like skin colour, but the connection is alot less profound and categorical than most people think.
high summer, not hug summer.
Wow, more than 2.4M of assembled bacteria in the new release of ABT! We plan to index these using our efficient colored De Bruijn graph index, Fulgor. We recently conducted experiments with nearly 1M genomes…getting there :) www.biorxiv.org/content/10.1...
I know England - Britain probably - needed the rain (gardeners and farmers breathing a sigh of relief) but it does feel like we’ve gone from hug summer to mid autumn in under 24 hours …
Wavefunction Collapse xkcd.com/3134/
![4-panel Comic. (1) STUDENT: If the wavefunction only collapses when I observe it, does that mean my consciousness affects the universe? (2) [Bad:] PROFESSOR: Yes, quantum entanglement proves that we all have souls. (3) [Good:] PROFESSOR: No, consciousness plays no role here. It’s just physical measurement. (4) [Chaotic:] PROFESSOR: No. The wavefunction collapses when *I* look at it because I’m a full professor. It won’t collapse for an undergraduate.](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:cz73r7iyiqn26upot4jtjdhk/bafkreic2mhmdvsftki6qe275hycolgt22msekbwzdyqoujt56qzpcbm3hm@jpeg "Click to view full size")
The medaka paper was a tour de force lead by two great students: @saulpierotti.bsky.social in my research group and Bettina Welz in @wittbrodtlab.bsky.social's group.
Similarly discovering 2-loci GxG effects - which are as rare as hen's teeth in human GWAS (though see above on our OCT study) shows us new biology; like all genetics, low variance of genetic variants does not imply low impact if drugged (cf statins and HMG-CoA Reductase)
Finally of course discovering GxE effects gives us new insights into environmental impacts - and sometimes those environmental impacts are easier to change than the genetic basis of course (eg, stopping smoking, or stopping drinking)
However, in the scenario of overdominance (the hetrezygote mean is above or below one of the homozygote means) - impossible to get for additive models - the additive beta estimate becomes an unstable, frequency dependent parameter. This is ... not healthy for on-going analysis.
Another aspect is that we often use additive parameters, eg, the slope of the association, often called beta, in meta-analysis or other analyses, such as Mendelian Randomisation. The assumption is that this slope (beta) is well estimated and slope by GWAS
Interestingly in our own work on common variants to OCT based retinal phenotypes we could see some evidence of 2 loci GxG and also association with a rare gene locus with variable penetrance journals.plos.org/plosgenetics...
Furthermore you will take into account GxE, though this is complex because we rarely know these terms - this looks like variable penetrance to us.
Sort of obviously in rare disease diagnosis - where you want v v low false positives and you are conditioning on the alleles in one person, you don't just say "oh it will be fine to use an additive model" - obviously you take into account dominance/recessive if present;
However, the fact that additivity is... fine... for discovery, and probably also out of sample population assessment (PRS etc) doesn't mean it is good for everything we want to do.
As @mikeinouye.bsky.social says, this really shows that the "mainly additive" observation in human is completely consistent with the "pervasive non additivity" seen in model organisms and farm animals - it is allele frequence and model estimation of small proportion of variance things
Unsurprisingly to actually discover and estimate GxE and dominance / GxG interactions you need either way more samples (10 million or more, not 0.5 million) or far higher allele frequencies.
Similar to the work you cite here (Hill etc) and your own work, we also agree that most non-additive containing loci are adequately discovered using an additive model. Two provisos though - having the causal variant typed *really* helps+you have the measure the relevant environments reasonably well
We then took all the effect size estimates of addivity, dominance, GxE and GxE to a simulation in a fully outbred setting - as we have in human cohorts - to ask with these genetic parameters what would we expect to discover / estimate
This is all in the context of wild-derived haplotypes (same "trick" of inbreeding from the wild Trudy Mackay does) and we can show that the alleles we find in our experimental crosses are present in the wild (and so would be part of a GWAS etc if we did this in Medaka)
We see lots of non-additivity - both GxE and GxG and we see third order terms - GxGxE and for loci we can experimental test (via CRISPR) that our statistical inference was right - that indeed we have the causal allele, it does have GxE; we can show genetic background effects as well - GxG
Nice blog and good to see this also from the twins/shared environment side. We (with my colleagues in @wittbrodtlab.bsky.social) have tried to tackle the non-additive in experimental settings (in medaka fish) which we can map to human (as the medaka fish are "wild") www.biorxiv.org/content/10.1...
I wrote about gene-gene interactions (epistasis) and the implications for heritability, trait definitions, natural selection, and therapeutic interventions. Biology is clearly full of causal interactions, so why don't we see them in the data? A 🧵:
What utter horror is this?
Oh dear. Not good. Epigenetics … facecream ? What does it mean?
Good to see @bsky.app has emerged as the social media platform of choice for science stuff arstechnica.com/science/2025...
:) there is an interesting question about hype titration - cheekily, what has been a scientific breakthrough at the same level as AlphaFold that had "appropriate" hype level?
Ultimately the dissallusionment going on is mainly focused on LLMs being somehow generic - they are not (or at least not now, but I think this is deep) - they are language machines. Oh - translation is a good thing, but often 80% solution not 99% solution.
To note "generative models" have been around certainly since the 1960s (hidden Markov models etc, Kalman filters - first use cases in the 1960s - passive sonar and cruise missile guidance respectively) so it is not the generative nature of them that is different.
Sure; this goes to what is AGI or some sort of AI assisstant anyway. Note: the image segmentation stuff is not quite "put into excel" but it is getting close (ie, a rather generic task that is very complex and labour intensive, and the AI system does it well with these days less training data)
There is more than alphafold in science. Most obviously there has been a complete revolution in image analysis
1st Bluesky post with recent updates: 1/3 It was incredible to attend and speak at the ReNU Hope Conference in Long Island, NY. This was the first time I met the families and the ReNU warriors, sharing many touching stories. I hope to continue advocacy in China where few are currently diagnosed.
Alphafold, chromebpNet, numerous image analysis - segmentation (everything from protein complexes to moving hearts) and representation learning, many other bits science
The (semi) wilds of Northumberland in August - sun burning through hill mist, dew still on the grass, cows and sheep calling out, loud birdsong on top, butterflies labouring between wild flower blooms - a million miles from London and Cambridge
All the chargers working at full whack - its MW draw down. I wonder if the grid will need to get predictive of fast charge draw downs
This actually meant we had enough to finish before we had bought our supper and got our tea :)
Heading north for the last weekend of summer; forced by bad EV infrastructure further south to make our first charge at Weatherby (top tip - use the newer chargers around the corner) - with near empty battery got full 72kW charge :)
London apple crop is looking good
So - just a reminder when the headlines shift on us (not sure how much they will, but I bet they will) that AI / Deep Learning is ... here to stay, at least for molecular, cellular and organismal biology!
Third, stepping back, these things always go in cycles. Some froth removal happens - both in academic enthusiasm and in commercial applications - as understandably people speculate on what will work (again, both academically and commercially)
Second, AI / DeepLearning has also become utterly critical and almost routine in image analysis - a lot of biology (my broader field) is about image analysis and we're never going to somehow throw away this useful tool. Segmentation, representation learning, autoencoders - all great.
(it is worth noting that the "clean" problem - just from a single amino acid sequence, and not leveraging evolution - is not solved; in the 1990s using evolutionary was absolutely brought in, so it is not that AlphaFold was the first to do this, but there is something ... still to be done here!)
First off, AI (or Deep Learning - I can forsee people going back to Deep Learning if AI becomes too jaded) has solved some "grand challenge" problems - AlphaFold and the folding problem is never going to get old for this - it was a grand challenge problem from the 1970s; solved "well enough" 2020
some more musing on AI - first off, I am not sure if there is a full blown AI winter coming back, or there is "just" some high winds and storms to blow away some of the froth. In the perhaps inevitable disillusionment with Large Language Models some things to remember:
Wow. That's dense. It feels to me that there is alot of body structure/body size correlations - is there anyway to screen those out (naive approach #1 - provide as covariate?)
This is .... cool.
⏳ One week left to register! Join us at EuroBioC2025 in Barcelona and the Bioconductor Carpentry Workshops for hands-on learning, community building, and cutting-edge bioinformatics. 🗓️ Registration closes August 28 — don’t miss out! 🔗 eurobioc2025.bioconductor.org #EuroBioC2025 #RStats
I think the contrast to post translation modifications and indeed phosphorylation would show ... many more than 11 publications!
But lumping some (but not others) into epi-XXXXX and then different people slicing up this rich complex world with different definitions of epi-XXXXX just. Doesn't help communication.
And it involves - mitotic cell state stability - and mitotic cell state changes - and all of development (!) - and of course X inactivation - and non-DNA based inheritance both inter-generational and trans-generational (when that happens). Not to mention cancer, where it is big.
And where would biology be without phospho/acetyl/ubiquin etc groups on proteins ... and we've got a whole new world (cue music) on RNA modifications. All this decoration and involvement is great and truly worthy of study (some from my group!)
I *love* work on DNA modifications and think they are very important; I *love* work on imprinting and think it is cool. X inactivation - bring it on!. Histone modifcations - hold the front page - they are important and a great state read out. non coding RNA rock!
Grrrr. We don't call protein phosphorylation or acteylation epiproteomics. So let's not call RNA modification epitranscriptomics. In fact, realisitically, epigenetics has so many definitions, some of them non-overlapping, that this whole epi-XXXX (genetics/genomics whatever) is just not helping us.
🚨 Imagine a bacterium that refuses to follow the textbook: It grows as tangled filaments, divides unevenly, reshapes its own membranes… and even builds grappling hooks. Meet Litorilinea aerophila — and here’s why it blew our minds.
Got a number in mind? A new pre-print by @lshtm.bsky.social colleagues surveyed 13,000 participants, and found it's about 9 per day in the current era. It also looks at how social mixing varies with age, ethnicity and social economic status. Link to paper:
Hey everybody! @drjuliawester.bsky.social and I have a new paper! We surveyed over 800 scientists, science communicators, and science educators who use social media. Conclusion: Scientists no longer find Twitter useful or pleasant, and many have switched to Bluesky! 🧪🌎🦑 doi.org/10.1093/icb/...
It is great. I spent a happy day today exploring some fresh GWAS. Love it!
(PS I know I am howling in the wind here. And - to my epidemiologist friends - I would like to apologise in advance for the nails-on-the-blackboard moment every time you here an AI person say "causal mask" or something like that)
Using the phrase "causal mask" about future not being able to influence the past is ... not a good phrase. Antecedent mask is far better.
Causal variables have to be antecedents to the variables they cause, but not all correlated antecedents have to be causal; indeed in observational biology/medicine most/many (depending on your set up!) are not. Welcome to ... the real world.
Causal has some stronger (and useful!) implications, namely that if you could change this variable this other variable would change if one could ... do the experiment again or live this little time series again (ie, one can talk about causality even when it is really counterfactual ideas)
However, AI folk, it is going to be far, far better if we collectively call "back in time" or "left" in the normal way to lay out time as "antecedent" and *not* causal.
Just had a very informative conversation / whiteboard session with my postdoc about attention network segregation on AI stuff. My "emotional understanding" of AI is ... deepening.
Oh - special mention for the King’s Cross tunnel system (seemingly signposted to ensure tourists get a work out); my map extends to Paris Gare du Nord / Gare de l’Est (but not yet Gare du Lyon - need google maps for that though I know it is easy!)
There’s some serious neuronal energy put into this and as I’ve moved where I’ve lived around London more and more islands “merge”
Then there is sort of space dilation on the edges (metropolitan line, north west; Piccadilly line both ends)
Mostly this is deformations (eg: Queensway / Bayswater) but some it is whole scale cut and change (South Tottenham / Tottenham Hale) and some you have factor in the platform and tube station layout (Bank/Embankment; Liverpool Street / Moorgate with the Elizabeth line)
When the islands merge they rarely honour the snap-to-grid and distortions of the tube map and then I can feel myself mentally moving bits of the tube map to keep the directions consistent