Lightnews — Scholar-powered news

Reposted by Michel Nivard

Ewan Birney @ewanbirney.bsky.social · 3h

I am hiring! - looking for a Staff Scientist to co-run my research group with me. Staff Scientist is a senior professional scientist role at EMBL. Please forward to people you might know who could be interested! embl.wd103.myworkdayjobs.com/en-US/EMBL/j...

Staff Scientist

About EMBL-EBI EMBL’s European Bioinformatics Institute is a data powerhouse, utilised on a global scale to advance scientific discovery through bioinformatics and solutions to some of the world’s mos...

embl.wd103.myworkdayjobs.com

1 36 25

Michel Nivard @michelnivard.bsky.social · 15h

I don’t know where you are globally but their still touring with this album for a few months…

Michel Nivard @michelnivard.bsky.social · 20h

should possibly try to simulate the computer environment as closely as possible in a VM (or container)…

1

Michel Nivard @michelnivard.bsky.social · 20h

I was looking at my future health as well, that’s certainly more usefull…

1

Michel Nivard @michelnivard.bsky.social · 1d

Anyway I’ll have something online soon-ish

1 3

Michel Nivard @michelnivard.bsky.social · 1d

I am aiming to do a little better by retaining approximate marginal distributions and correlations, and as I said if I get really ambitious I might even generate some genotypes?

1 1

Michel Nivard @michelnivard.bsky.social · 1d

This isn’t mine, but I saw it l they say: “The datasets are therefore not internally consistent – for instance they may contain reports of prostate cancer linked to female participants, medical events after death or dates of disease without corresponding diagnoses.”

1

Reposted by Michel Nivard

Michel Nivard @michelnivard.bsky.social · 1d

Synthetic non-parametric UKB phenotypes straight from an R package. It pulls all the required summary data from the UKB website.. Somewhat tempted to go totally overboad and have it simulate synthetic genomes, with plausible SNP h2's and "real" top hits (which i'd pull from GWAS cat & nealelab)..

3 3 12

Reposted by Michel Nivard

Ewan Birney @ewanbirney.bsky.social · 1d

A reminder that the PhD program for EMBL is open for winter recruitment round. If you want to understand life at a molecular level and have a quantitative science (you don't have to have done biology - we love our mathematicians/physcists/chemists!) apply www.embl.org/about/info/e...

Application – EMBL International PhD Programme

www.embl.org

1 43 33

Michel Nivard @michelnivard.bsky.social · 1d

the Synthetic Kingdom of Great Britain and Northern Ireland Biobank sounds a little subversive somehow...

2

Michel Nivard @michelnivard.bsky.social · 1d

UK Synthbank? UK Simulbank? Synthetic Kingdom Biobank? (to be VERY clear this is for people to trial there methods pipelines, not one of those insane ideas of relacing real data with synthetics...)

Michel Nivard @michelnivard.bsky.social · 1d

Synthetic non-parametric UKB phenotypes straight from an R package. It pulls all the required summary data from the UKB website.. Somewhat tempted to go totally overboad and have it simulate synthetic genomes, with plausible SNP h2's and "real" top hits (which i'd pull from GWAS cat & nealelab)..

1 2

Michel Nivard @michelnivard.bsky.social · 1d

Actually it is parametric, but it cycles agnostivall trough a bunch (mixture) models making sure the quantiles and extreme's match...

1

Michel Nivard @michelnivard.bsky.social · 1d

Synthetic non-parametric UKB phenotypes straight from an R package. It pulls all the required summary data from the UKB website.. Somewhat tempted to go totally overboad and have it simulate synthetic genomes, with plausible SNP h2's and "real" top hits (which i'd pull from GWAS cat & nealelab)..

3 3 12

Michel Nivard @michelnivard.bsky.social · 2d

I am thinking ppl can use it to test code, pipelines etc. @dsquintana.bsky.social didnt you have a syth data package I could potentially pipe this into so I dont have to rebuild the core mechanics?

2 3

Michel Nivard @michelnivard.bsky.social · 2d

QUESTION 🚨: I wrote code that uses methods of moments too generate realistic synthetic data for a UKB trait from the showcase page. I might be able too add corelation between traits (have to find a source) to allow multivariate synth data, is this usefull enough to wrap in an R package??

2 1 7

Michel Nivard @michelnivard.bsky.social · 4d

reducing it to its most absurd could be an amazing way to highlight the silly...

4

Michel Nivard @michelnivard.bsky.social · 4d

I am in Berlin near the Max planck(s?) for a conference till thursday, LMK if any of you in the bluesky science-o-sphere want to meet (very limited availability, not cos I am cool and all booked up but just tired and old).

1 3 13

Michel Nivard @michelnivard.bsky.social · 4d

I was thinking Sam Altman’s Dutch Alt, “Simon Oudemans”… why aim low with Nature I was thinking NEJM…

4

Michel Nivard @michelnivard.bsky.social · 4d

I want to write this as a troll now… prob best under pseudonym…

1 3

Michel Nivard @michelnivard.bsky.social · 4d

someone is going to do the perverted inverse and try to replace RCTs with LLMs: “we prompted llama 3.3 71b with: ‘you have just taken a semaglutide inection; outline your meals for this week’ and proceeded to compute weight change as 1 kg loss/gain per 7000kcal under daily recommended intake”

2 1 8

Michel Nivard @michelnivard.bsky.social · 4d

It was a glorious 6 weeks for Old Navy Dads..

1 2

Michel Nivard @michelnivard.bsky.social · 4d

There is that quote about religion being the opium of the masses? (I am surely bastardizing it here…)

1 1

Michel Nivard @michelnivard.bsky.social · 4d

I was here with a 6 and 8 year old so they sprinted past and I had no time to dig into the what/why/how…

1

Michel Nivard @michelnivard.bsky.social · 5d

There is a Dutch museum about all religions, they also cover post religious pseudo religious play, including Warhammer (?) (cc @lastpositivist.bsky.social warhammer as post religion cosplay, could be a thesis is all I am saying..)

4 2 25

Michel Nivard @michelnivard.bsky.social · 6d

I have even seen really creative work using generative models. This is just about the dumbest idea possible…

1