Simon Gravel
@SFGravel
Followers
1,386
Following
92
Media
15
Statuses
394
Population genetics at McGill university. @SFGravel @ecoevo .social
Joined August 2012
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Eagles
• 230970 Tweets
Dick Cheney
• 192590 Tweets
Packers
• 130317 Tweets
ENGFA ACTRESS 100M
• 118238 Tweets
श्री गणेश
• 113163 Tweets
Ganesh Chaturthi
• 103865 Tweets
Ganesh Chaturthi
• 103865 Tweets
गणपति बप्पा
• 81210 Tweets
Saquon
• 76054 Tweets
Jalen
• 58603 Tweets
Barkley
• 44004 Tweets
भगवान गणेश
• 41183 Tweets
#二度と撮れない画像を貼れ
• 34726 Tweets
Jordan Love
• 32308 Tweets
#GanpatiBappaMorya
• 28779 Tweets
Party Love AndaLookkaew
• 26724 Tweets
JustinOn NCAA100Kickoff
• 25401 Tweets
ビッグラン
• 15478 Tweets
イコラブ
• 12766 Tweets
#学マスデビュー初心_名古屋
• 11232 Tweets
Nikocado Avocado
• 10555 Tweets
Many papers reported genetic evidence for 'ghost' archaic admixture in early human history. With new WGS data from Nama (Khoe-San) & other African individuals and parameterized modelling, we (
@apragsdale
@HennLab
et al) reach different conclusions.
7
81
214
A lot of genetic studies discard data from minority groups. But diversity without (statistical) inclusion is exclusion.
It’s hard to agree on whether and when it is ok to discard data, but I think we can do better by being more thoughtful about it.
3
67
145
A figure in the All of Us paper was criticized on many fronts, including the use of a dimension reduction technique (UMAP). Some people concluded that UMAP is not appropriate for representing genetic ancestry. I disagree.
@jkpritch
@ras_nielsen
@dgmacarthur
1/n
5
26
142
Lots of interesting recent methods & results on intricate models of human genetic history. We're having a mini-symposium next week May 6-7 to follow up on discussions at
#probgen21
.
Free registration:
4
56
108
Submission rejected at
@biorxivpreprint
because it included a French version of the abstract (preprint, in English, is on French Canadian genetic history).
Only English allowed.
Are you afraid I'll sneak in offensive material in the French abstract?
6
17
99
Shadi Zabad’s latest work on predicting genetic risk:
Variational Inference for Polygenic Risk Score (VIPRS 🐍) consistently competes with or outperforms leading approaches.
1
32
84
It may be time to retire or resequence some pioneering sequencing datasets. (With
@LukeAnderTroc
)
4
35
73
Coalescent theory allows for powerful simulations (eg
@jeromekelleher
’s msprime), but is biased for large sample sizes. Student D. Nelson et al. found a fix for large biases in IBD and ancestry distributions, expanding on an idea of Bhaskar et al 2014.
3
30
73
A new tenure-track position in computational genomics at McGill:
0
99
73
If you like UMAP, you'll love this preprint allowing for much easier interpretation of this structure you've been staring at.
If you hate UMAP, you'll love that this preprint formalizes a lot of hand-wavy interpretations of UMAP plots. (Ok, maybe not love it. Appreciate?)
Out now! We study genetic structure in large biobanks using topological data analysis via UMAP and HDBSCAN. This approach is fast, easy-to-use, fits into existing pipelines, uses data you already have, and is downright fascinating. 1/
3
29
128
1
18
60
Our attempt at explaining peer review to kids -- now in English.
3
10
58
Mini-workshop next week, please RT!
------------
Adaptation in structured populations
Thursday, June 29
------------
Featuring
@apragsdale
@AmyAGoldberg
@jblanc12
@QuintanaMurci
@nanditagarud
.
Free registration:
Program overview:
1
57
57
Our study on the history, genetics, and mathematics of the Great Migration of African-Americans
@PLOSGenetics
3
30
50
Presenting VIPRS, a Variational Inference (VI) approach to polygenic risk score estimation. .
PhD student Shadi Zabad shows that adapting a fine mapping approach to scale genome-wide gives fast, robust, and accurate risk prediction. (with Yue Li)
3
15
50
We want more diversity in genomics cohorts. But how do we address the real issues in community engagement, cohort hetereogeneity, and statistical modelling? We invite researchers, funders, and the community to discuss June 15-17, 2018 in Montreal, Canada
0
37
49
@apragsdale
's paper is out in Nature!
Coverage by
@carlzimmer
@NYTScience
:
And a perspective by
@DrEleanorScerri
:
Here's my original explainer:
Many papers reported genetic evidence for 'ghost' archaic admixture in early human history. With new WGS data from Nama (Khoe-San) & other African individuals and parameterized modelling, we (
@apragsdale
@HennLab
et al) reach different conclusions.
7
81
214
6
19
42
Preprint by postdoc
@i_krukov
: .
Most methods for modelling distributions of allele frequencies rely on either weak selection or small sample size assumptions. Ivan shows how to jointly handle large samples and strong selection.
1
18
41
We find that contemporary human population structure in Africa only dates back to 120-135ka. Prior to ~120ka, under a weakly structured stem, the human population history is highly reticulated.
3
4
37
Also, Dominic Nelson's Wright Fisher extension to
@jeromekelleher
's msprime is just out in PLoS Genetics
. Stay tuned for more large scale popgen simulations.
1
19
40
McGill offers new PhD program in Quantitative Life Sciences.
Train in life sciences with peers and instructors serious about advanced math, stats, and cs! Yes, that includes deep learning as well.
0
29
40
Looking for resources to help students organize comp bio projects. Best practices in data organization, reproducible code, docs, collabs. Any favourites?
Got good mileage from , but looking for more recent alternatives (incl. version control).
1
10
36
If we are correct, adaptation in early human history would differ from what we'd expect following admixture with a deeply diverged hominin.
Morphologically divergent fossils, such as Homo naledi, would be unlikely to represent branches that contributed much to Homo sapiens.
1
1
33
@LukeAnderTroc
explored the appearance of population structure in Quebec over centuries in using thousands of genomes and millions of genealogical records! . Story involves rivers, mountains, Royal hunting grounds, and asteroids.
3
10
31
Rather than archaic admixture in Africa, our best supported model has migration among closely related populations going back hundreds of thousands of years.
This deep but weakly structured human (hominin) stem has only modest effect on present-day population differentiation.
1
1
31
Congrats to
@LukeAnderTroc
who successfully defended his PhD thesis today! Here he discusses the impact of an ancient meteorite on genetic diversity in Québec.
3
2
32
Dominic Nelson's heroic inference of allele transmissions in very (very) large genealogies:
2
13
30
French version of this tweet:
Notre soumission a été refusée chez
@biorxivpreprint
à cause d'un résumé en français, cibole. Y veulent yenk de l'anglais, baswell. Chu t'en tabarnouche.
1
1
30
@apragsdale
heroically validated this model against multiple lines of genetic evidence that had been previously used to argue for archaic admixture, including LD statistics, the conditional SFS, and cross-coalescence rates. The model does well for all.
1
1
25
Making mistakes in genetic simulations is all too easy. Some suggestions for users, developers, and the entire community to make research results more robust.
Do you use msprime? You'll want to check out this commentary in our latest issue "Lessons Learned from Bugs in Models of Human History"
@apragsdale
@SFGravel
@jeromekelleher
1
7
14
0
7
24
Links to the papers: , ,
0
3
25
Shadi Zabad shows that a common (and unnecessary) assumption relating allele frequency and effect size in LD score regression underestimates functional enrichment. .
Stratifying by allele frequency helps reduce bias but is no magic bullet. (1/3)
1
7
23
IRB approvals do not exempt scientists from thinking about ethical consequences of research, and from speaking up about problematic work. IRBs rely on info provided by research proponents, and can make errors. IRB approval is necessary, not sufficient.
0
3
22
Congrats to postdoc
@apragsdale
who will be joining the faculty at UW Madison
@UWiBio
! Also, congrats to
@UWiBio
for hiring such an outstanding scholar! Working with Aaron has been a blast!
1
0
22
Lots of positions in Montreal for running+analyzing biobanks with -omics data. Cartagene is hiring
- A data curator
- An operations manager
- A biostatistician
And I'll have a few positions very soon for postdoc+students in stat and population genetics!
1
10
22
PCA flattens data and puts different populations on top of each other. Admixture plots assume that there are “true” type populations that mix in different proportions to create individuals. UMAP massively distorts distances to show each data point and preserve neighborhoods. 3/n
1
5
21
Last week to submit posters that address representation issues in genomics: we welcome approaches from statistical & population -omics, community engagement, ethics, study design, etc.
Closing the genomics research gap
June 15-17, 2018
Montreal, Canada
1
14
19
McGill's Genome Centre has many new TT faculty positions in q-bio, statistical and population genetics. .
0
42
20
Open rank search at
@mcgillu
in genomic medicine: .
A rare opportunity with enough resources to build computational + experimental programs!
0
12
19
UMAP is appealing because it can reveal patterns in the data that would not have been obvious otherwise. By contrast with PCA or admixture, it can reveal multiple levels of discrete and continuous population structure in one plot. See
@adp_diaz
’s papers & thesis on the topic. 5/n
3
5
19
Still amazed by how useful the demes package is for specifying demographic models ().
Testing competing models in minutes instead of weeks never gets old.
0
4
17
UMAP is used to get a visual representation of complex (high-dimensional) data. To do this, it distorts the data. A lot. This is unavoidable: genetic data is high dimensional, and any plot that fits on a piece of paper has the potential to hide and mislead. 2/n
2
2
16
UMAP gives roughly equal visual space for each participant. This makes UMAP distances meaningless, but also makes UMAP useful to visualize the composition of a cohort. A large UMAP feature includes many individuals. This is useful information. 6/n
3
1
17
Ahead of discussions+training next week at
@McGillGenome
about equity in genomics research training, I wanted to share these useful
@UCSF
guides for trainees and profs on talking about race and inequality in science.
0
5
17
Thanks to Soheil for helping wrap this up from his new industry position!
On the decidability of population size histories from finite allele frequency spectra
0
9
16
@ProfLikeSubst
Given that some PIs actively discourage trainees from having kids and claim that they are career-killers, it's fine to have some researchers point out that having kids is ok and can even be nice. Obviously, shouldn't be directions or unsolicited advice to their trainees (please!)
1
0
15
Lots of other cool finalists at this year's
#ScienceExposed
data viz competition, but
@adp_diaz
's UMAP figure is the only non-imaging one -- support your fellow data nerd and vote :)
1
3
14
Tenure-track position at McGill in statistical genetics:
http://t.co/Rw9uxBMcu7
It's an awesome place to do stat/popgen research!
0
50
14
A CERC-funded faculty position opening in population genomics at McGill!
0
13
14
Sail on, sail on
Oh mighty ship of State
To the shores of need
Past the reefs of greed
Through the Squalls of hate
-Leonard Cohen, Democracy
0
5
14
The most useful method depends on what we want to do. UMAP has become popular to summarize cohorts. Critics have lamented this, arguing that the plots are pretty but misleading. They can be both, but presumably that is not the main reason they are popular. 4/n
1
1
13
A UMAP feature does not mean something is real or relevant. But showing a UMAP plot and discussing what we see is a good way to identify relevant structure in the data. Labeled properly, it emphasizes continental groupings less than most PCA or admixture plots. 7/n
1
1
12
Our latest preprint: The Great Migration and African-American genomic diversity
http://t.co/Jnw1PuEQsf.
Congrats to postdoc Soheil Baharian!
0
15
12
Our latest work on the frequency spectrum with new software, new math, new applications, and ploids!
1
7
11
Last week to submit abstracts for RECOMB-genetics! New abstract deadline is Feb 16.
RECOMB-genetics will take place on April 19-20 in Paris just ahead of the main RECOMB meeting (April 21-24), with focus on genetics x cs x stats (+ popgen of course).
1
15
11
After years of retweeting preprints based on title & authors, or (if I am being scholarly) abstracts, twitter finally saw through my little game.
1
0
10
Abstract deadline now February 13 for Recomb-Genetics (May 4, 2019, Washington DC). Keynote speakers Sharon Browning and Josh Akey, ahead of
@recomb2019
featuring
@cdbustamante
,
@Alfons_Valencia
et al.
Expect haplotypes and ancient history!
Please RT :)
0
10
10
I join
@LukeAnderTroc
in thanking our co-authors
including Dominic Nelson, Shadi Zabad,
@adp_diaz
@jeromekelleher
, Ivan Krukov,
@benjeffery
@genotepes
,
the teams at
@_CARTaGENE_
and
@BALSAC_UQAC
, and (43,000 times) the generous participants at
@_CARTaGENE_
1
0
9
Those of you at
#SMBE2019
should check out
@apragsdale
's talk on Wednesday! Heroic work including theory, software, and applications.
Excited to be in Manchester for SMBE - meeting old friends and new, and seeing what you all have been up to! I'm speaking Weds at 3:30 about using multi-population LD models to learn about deep events in human history (incl our recent paper in plos gen: )
1
6
46
0
2
9
Ten days left to submit abstracts for Recomb-Genetics (May 4, 2019, Washington DC).
Guest speakers Sharon Browning and Josh Akey, just ahead of
@recomb2019
featuring
@cdbustamante
,
@Alfons_Valencia
et al.
Expect haplotypes and ancient history!
0
7
9
Beau travail de
@LukeAnderTroc
pour visualiser
#covid
avec les données de
@OpenGovON
. Pourrait-on avoir les mêmes données à
@sante_qc
?
I wrote an
#opensource
script that automatically downloads and plots the latest
#COVID19
data from
#Ontario
. Scripts are available here :
#DataVisualization
#DataScience
#covid19Canada
#ontarioshutdown
1
13
42
1
2
9
@leland_mcinnes
@adp_diaz
We ran the
@1000genomes
project data without PC filtering, and it looked very good
0
2
9
Our latest preprint on simulating allele frequency distributions: . Congrats to Julien Jouganous! Comments welcome.
0
2
7
A key point of our article is that individuals who self-identify as African-Americans do not form a genetically homogeneous population. 1/3
1
4
8
Thanks
@HennLab
! It was great to catch up! I may want to use the top right photo as cover art for
@LukeAnderTroc
's next manuscript...
0
0
7
@kzshabazz
@monicaMedHist
@SandyDarity
Ancestry+demography (not race) are important to understand genetic diversity
2
8
8
@LukeAnderTroc
describes batch effects in the 1000G project (and their consequences) .
You may want to check your favourite imputed SNPs against his list of batchy SNPS,
,
Also check out those of
@FabriMafe
!
1
5
8
@michaelhoffman
I should have reserved that domain! My favorite gravel sign so far still gets regular use on my office door:
1
0
7
@adp_diaz
' approach using
@leland_mcinnes
and Healy's UMAP applied to GnomAD data
(UMAP:
Alex's paper )
6
46
140
0
1
6
Therefore genetics research should not stop at arbitrary boundaries set by a racist society. 3/3
@monicaMedHist
@kzshabazz
@SandyDarity
5
6
6
Take home: If you are excited about distant Native ancestry in your family, that’s great, but avoid overplaying it. If you have overplayed it in the past, you can acknowledge the mistake (I have made that mistake!) without disavowing your ancestors. 4/4
0
0
6
With talks by
@apragsdale
@sr_sankararaman
&Erin Molloy, Paul Verdu,
@HennLab
@morestrada
, and
@QuintanaMurci
0
0
6
Marketing emails from ASHG vendors are off the chart this year. Both in terms of volume and spamminess. Assuming some email list leaked?
@GeneticsSociety
, any chance you can rein in vendors?
0
0
6
@sbmontgom
@marcus_NJ
@popgengoogling
@AwadallaLab
@ArmandeAngHoule
@hgibling
@KimSkead
As a French Canadian, I would never joke about poutine. It's clearly an outsider.
0
0
6
PhD and postdoc positions at McGill University. Involves genetics, math, history, biology, computing, and cool data.
http://t.co/tbhCMhCMq4
0
12
5
We were also thinking of funding this project through the sale of art prints, but then I'm not sure we can use
@uk_biobank
data for commercial purposes :)
0
0
5
The UMAP method was developed by
@leland_mcinnes
and John Healy for visualizing high-dimensional data (). Each dot is an individual in the UK biobank, coloured by self-identified ancestry categories ( )
0
0
1
1
0
6
Les résultats du sondage sur la liberté académique peuvent donner à penser que le monde universitaire est devenu incroyablement timide. 60% des professeurs s’autocensurent en évitant certains mots! 35% évitent même certains sujets! Est-ce la dictature du politiquement correct?
1
1
6
@adamauton
@rdhernand
@LukeAnderTroc
Someone should analyze the AFS (acronym frequency spectrum) for the 1000GP/TGP/kGP
0
1
6
Population history shapes genetic diversity at all levels! There are differences in ancestry at continental, regional, and local levels. 2/3
1
3
6
I like the approach of this paper:
1
2
6
Tests génétiques: attention aux tests bidon! Aucun test d’origine génétique ne donne de statut légal! via
@lp_lapresse
0
3
5
This is starting in 30 minutes!
Lots of interesting recent methods & results on intricate models of human genetic history. We're having a mini-symposium next week May 6-7 to follow up on discussions at
#probgen21
.
Free registration:
4
56
108
0
2
5
A beautiful news story on the Silent Genomes Project (in French)
0
1
9
@lu_sichu
Race is a socially defined label that is correlated with ancestry. So if you label individuals by race on a dimensionally reduced dataset, you will see patterns. But umap will also show variation, say, among French Canadians living across Quebec.
1
0
4
Surprisingly, large samples make it easier to handle selection. In large samples, the number of relevant ancestral lineages tends to be much smaller than the sample size. We can mathematically sacrifice the extra ancestors to model natural selection.
1
0
5
In short, there can be valid reasons to exclude data, but incentives of publication also tend to discourage more inclusive analyses. Some small changes to the way we analyze and report data could help foster inclusion without imposing an unreasonable burden on researchers.
0
0
5
@rdhernand
Or it's all coding but French is more robust to errors!
So CIHR could check whether scores for applications in French have lower variance across reviewers :)
2
0
5