Maia Munteanu

DNA mismatch repair (MMR) deficiency is an important driver of cancer genome instability. Yet it is usually treated as a binary label 0⃣/1⃣ MSI-high or not. Our new preprint – fresh from the oven🔥 -- argues that this simple dichotomy obscures the underlying biology. Trouble with different parts of the MMR machinery is "recorded" in the genome sequence as distinct mutational patterns🧬. We worked out which pattern means what. Amazing collaborative work led by Marcel McCullough and Adam Poti, with @maiamontagnarde Maia Munteanu, at @IRBBarcelona + @UCPH_Research Bravo, team!

@RuxandraTeslo @Heritage It probably also helps to reduce endometrial cancer risk, which tends to be higher in people with PCOS (probably due to several factors like unopposed estrogen, obesity, increased endometrial thickness, etc).

@hoffman_steven @stethospeaks The magrittr package introduced the pipe operator in R and The Treachery of Images is a very famous painting of a pipe

@maiamontagnarde @stethospeaks I don't get the Analogy for Magritte's are and what the magrittr R package does - not sure if this a a lack of understanding of the R package, the art or both

Can't believe that I had to visit the Art Institute of Chicago to see Georges Seurat's paintings in distinctive pointillism to realize that the single-cell RNA-seq analysis package in R is named Seurat because the appearance of cell clusters resembles the artist's style. 🫠

@CJHandmer @LauraForczyk There is actually a name for this: dyscalculia. Brains are plastic but also weird, some people genuinely cannot do mental arithmetic and it's separate from general cognitive abilities.

@LauraForczyk Obviously ymmv but brains are almost infinitely plastic and I think anyone can learn to do one significant figure quickly. Maybe only some people have the persistence to learn to get all 13 digits right.

I'm continually surprised by how rare mental arithmetic skill is among technical people. It's about as hard to learn as riding a bike (ie not hard) and it puts first principles thinking on warp drive. With daily practice you can quickly get faster results than using a calculator even for complicated operations. It also greatly improves physical intuition. I see it as a kind of literacy.

@ValerioCapraro That's not what regression towards the mean actually means though, it's a statistical sampling phenomenon not some fundamental societal risk.

Please read the following responses to the main critiques to this post: 1. Critique: I cannot reject a candidate because I believe their application is AI-generated, as it’s impossible to know for certain whether a text is AI-generated. It’s true that no algorithm can tell for sure if a text is AI-generated. However, we are discussing a specific piece of text, not a general one. The text contained dozens red flags, including words typically used by ChatGPT (intricate, delve…) and a suspicious change of font in the last sentence. Additionally, the statement continued in this manner for three pages, displaying red flag after red flag. The strongest indicator was the shape of the apostrophes: standard Calibri font has round apostrophes, GPT-generated text has straight ones. Moreover, the whole statement was superficial, lacking deep content; only form, no substance. At the end, I didn’t give it a zero - this was an exaggeration in my post to make a strong point, and I apologize for this - but I did give it a low grade, making it virtually impossible for the candidate to reach the interview. 2. Accusations of discrimination This accusation deeply hurt me and was likely the main reason I didn’t sleep all night. I have been politically active since I was a teenager, defending ideals of equity through volunteering and donations. Being accused of violating one of my foundational principles really hurt. Fortunately, it’s easy to address this criticism. I’m not discriminating against non-native English speakers for the simple reason that ALL our applicants are non-native English speakers. The accusation itself is just ridiculous. I’m simply giving a bad grade to a bad essay. Is this discrimination? 3. Exposing a student I take this point. I tried to prevent this criticism by removing personal information. However, I know that sharing private info in public is unjust. I take this point and I apologize for this. But, this should not obscure the real issue. The real problem that made so many people upset with my post isn’t that I exposed an anonymous individual. The problem is that I exposed a widespread behavior. In my work, I spend a significant amount of time reading essays at various levels, from research statements of PhD applicants to research articles. It’s alarming to see how many of these are written by generative AI. Millions of people use ChatGPT to replace their work. It’s alarming to see how many of these are written by generative AI. Millions of people use ChatGPT to replace their work. I personally use ChatGPT a lot. This post itself has traces of ChatGPT. But this isn’t a problem, as long as one uses ChatGPT to refine writing or suggest improvements. ChatGPT should be used for enhancing your work, not replacing it. Unfortunately, many people use ChatGPT to replace their work. This is bad for them (because they won’t improve) and to society overall, as we risk a regression to the mean. The message of my post was: don’t replace your unique contributions with average outputs. Be unique. I hope this clarifies my position. Thank you for reading.

TOMORROW ‼️ #Incubakers solidarity breakfast 🍽 to support the association @As_Dyrk1a_ES 🧬 📆 21 Mar ⏰ 9:00 - 11:00h 📍 #PRBB inner square ☕ + 🥧 = 3€ (BYOM) See you there #PRBBCommunity!

@Botanygeek Genuine question: isn't UPF considered to be less healthy because of the ultraprocessing of its ingredients rather than just the nutritional content itself? My understanding is that, even when nutritional content is controlled for, UPF is still worse than home-cooked food.

@AshaKrupka @strnr I find GPT4 to be pretty good at writing/debugging Nextflow code.

@strnr Good prototype, but probably far from being functional. For #snakemake ChatGPT doesn't get most of advanced functions. Unlike python or R, it's relatively useless for debugging workflow syntax as well.

I sketched out a typical variant calling pipeline, asked GPT-4 to read my chicken scratch handwriting and write a Nextflow pipeline to run the analysis 🤯

@ivyyy022 @fentpot @Aella_Girl Yup! If you’re studying one relationship between two variables, more data -> better coverage of the underlying population -> test is more informative. But in a large dataset with many variables and observations you’ll get a lot of spurious associations if you test everything.

@maiamontagnarde @fentpot @Aella_Girl If I’m following you correctly, a larger sample size make spurious correlations less likely, but a larger # of variables in a dataset makes them more likely. Is that right?

women with greater belief in the supernatural rate themselves as very slightly less hot than other people of their same demographic in general, men rate themselves as hotter compared to their demographic vs women overall (n=164,000, y-axis -3 to 3)

@fentpot @Aella_Girl @ivyyy022 Agreed! My comment was more to say: you can absolutely p-hack with a large dataset and it’s actually much easier than with a smaller one, so smaller p-values thresholds and correction are kinda necessary if testing stuff randomly.

@maiamontagnarde @Aella_Girl @ivyyy022 yes, in that with big data there are often more hypotheses to feasibly look at. but holding the number of hypotheses looked at constant, at a given significance level you aren’t more likely to find spurious positives with more data than less

@fentpot @Aella_Girl @ivyyy022 That’s true - if you have a hypothesis and you test that, you are more likely to detect a real relationship, if it exists. But if you’re exploring a dataset, with no particular hypothesis, you’re more likely to find statistically significant relationships where there are none.

@maiamontagnarde @Aella_Girl @ivyyy022 i know this is a common thing to say, but i am not sure i agree- if you decide your hypothesis beforehand, conditioning on the hypothesis being false, you are less likely to get a spurious low p-value positive with higher n than lower n, not the reverse

@Aella_Girl @ivyyy022 Large datasets usually generate much smaller p-values, so using the same p threshold for a dataset of 100 observations and 100k observations, will inevitably mean you find a lot more statistically significant relationships in the latter. doi.org/10.1287/isre.2…

@ivyyy022 it's not p-hacking if the sample size is sufficiently large!

@ChenxinLi2 @ATinyGreenCell Fyi Tropic is not colorblind-friendly (for deuteranopia and monochromacy).

@ATinyGreenCell This is the "Tropic" color scale from rcartocolor (github.com/Nowosad/rcarto…). Highly recommend this package for colors.

Spent an hour or so simulating some data. Left: f(x, y) with a global optimum; right f(x, y) with 4 local optima.

☢️Our space radiation/mutagenesis paper is out!☢️ Nice work Tiffany @tm_delhomme Maia @maiamontagnarde & Josep and importantly our collaborators at Columbia ⚛️RARAF⚛️ Manuela and Veljko! Read our study here 👇 nature.com/articles/s4159…

‼️ Apply to the #PhD student call at the @IRBBarcelona: ➡️exciting 🧪👩‍🔬& a great 🏙️ ➡️jumpstart your #reserarch career in #biomedicine ➡️only 4 days left to submit application! Please RT &/| forward to anyone interested 👇

‼️New preprint out‼️ Brought to you by the mighty Marina @msalvadoresf and yours truly 🚴‍♂️🔀🧬"Cell cycle alterations associate with a #redistribution of #mutation rates across chromosomal domains in human cancers" biorxiv.org/content/10.110… Here is a🧵! 1/ 👇

@damiendamien @TAndersen_nSCIr @GeorgeMonbiot Not necessarily, there are still population-wide genetic differences that can contribute to variation in height.

@TAndersen_nSCIr @GeorgeMonbiot UK & Netherlands are < 100miles apart. The genetics are the same.

#RegenesisFact 16: Probably as a result of a disastrous combination of extreme inequality and the failure to regulate the food industry effectively, 5-year old boys in the UK are on average shorter than those in any comparable nation. 🧵

Filled with new ideas and enthusiasm for #genomics research, we have returned from the Genome Data Lab scientific retreat. There was lots of cool #science, a collaborative spirit and great times at Salou/Tarragona!

A new #CancerPrevention vaccine could be coming, and Dr. Eduardo Vilar-Sanchez is leading the team developing it. The vaccine is for people with Lynch syndrome and it will soon be tested in #ClinicalTrials. go.usa.gov/xM64W @evilarsan #NothingWillStopUs 📸: @MDAndersonNews