How every layer of science's "self-correcting machinery" failed when Iva Veseli and I simply wanted to reproduce the findings of a high-profile study on gut microbiome and autism:
merenlab.org/2026/04/15/u...
That's what we saw. I guess it's a matter of how you define "better". At some point I think the gain in recall is worth it.
That's important and straightforward to incorporate to MAG-E. I'll follow up on this. Thanks
Absolute banger:
"find that metaSPAdes consistently outperforms MEGAHIT"
"Binning refinement, which combines bins from multiple different algorithms, leads to reduced performance"
"We further show that CheckM2 systematically overestimates completeness and underestimates contamination"
This simulation-cum-benchmark study on MAG making by @tkorem.bsky.social & team looks really interesting. Loads of plots and results to work through!
www.biorxiv.org/content/10.6...
We were also surprised! "Classic" inners work better with multi-sample (particularly CONCOCT!), but COMEbin and semibin2 are actually better in single-sample (at least in this benchmark and another unpublished one)
i've yet to see evidence that this man is anything other than the fucking coolest to ever do it
The truly visionary thing for Simons to do would be to map the collaboration graph of submitted proposals, find the bridges between large connected components, and offer them unrestricted awards
Schopenhauer
Pollution dropping by that much likely means a whole bunch of people are alive who otherwise wouldn’t be. Incredible stuff.
We just published in @molsystbiol.org with the Mugler lab (UPitt) on bacterial population dynamics during tumor colonization (mouse model). Our study was guided by a Luria–Delbrück-style idea: infer mechanism from statistics (1/7) 🧪🦠
doi.org/10.1038/s443...
Out after peer-review: www.science.org/doi/full/10....
Our bottom line stayed: never use leave-one-out cross-validation as it has inherent train-test leakage. Consider our Rebalanced version instead!
We now also account for regression and nested cross-validation, with more extensive benchmarking.
Using leave-one-out cross-validation to calculate metrics such as AUC and R^2 creates bias! This can be fixed by removing one of each class in the meanwhile to maintain the training data distribution - great work by Tal and his team 🧪🧬🖥️
I found a flowchart which helps you navigate the IT landscape
Back when we hid beer in the cold room in a box labeled “yeast embryos.”
Out after peer-review: www.science.org/doi/full/10....
Our bottom line stayed: never use leave-one-out cross-validation as it has inherent train-test leakage. Consider our Rebalanced version instead!
We now also account for regression and nested cross-validation, with more extensive benchmarking.
As long as three reviewers keep reading each proposal, doesn't really address anything either
"This could take down Democrats, too."
I know. And I frankly wouldn't give even an itty bitty damn if it implicated every Democratic man in Congress, every Democratic hopeful for 2028 and every Democrat who has even thought about running for office.
Down with the sex predators, wherever they are.
White paper: github.com/aqlaboratory...
GitHub Repo: github.com/aqlaboratory...
Huggingface: huggingface.co/OpenFold/Ope...
Testimonials: openfold.ghost.io/openfold3-co...
OpenFold3-preview (OF3p) is out: a sneak peek of our AF3-based structure prediction model. Our aim for OF3 is full AF3-parity for every modality. We now believe we have a clear path towards this goal and are releasing OF3p to enable building in the OF3 ecosystem. More👇
Just to be super clear, if you’re phoning in your peer review to ai you should quit your job so someone else who actually likes science can have it.
No idea. Still working through this.
This is from the very paper you linked to - Figure S5. They claim that this has a p-value of 1. It's not here and there, this is what most results look like, and this is a large part of the basis for claiming that there are no robust associations with other tumors.
So you look at this figure and your interpretation is "no signal"?
I never knew I needed this thread
We are hiring a postdoc - come work with us (www.koremlab.science) at the intersection of #microbiome, data science, and women's health!
Message or email me if interested. 🖥️ 🧬
I am seeking a postdoc for my group at UCLA. We work at the intersection of population genetics x microbiome (garud.eeb.ucla.edu). If interested, please message me!
This will also likely reduce the number of study sections, firing SROs and making them less specialized.
Hope this is useful - consensus statement "Guidelines for preventing and reporting contamination in low-biomass microbiome studies" rdcu.be/er3Io