2. Further, this next tutorial walks us through graphs & GNNs in an intuitive manner, while also going quite deep into the specific mathematical terminology of the field.
4. Building further, a good educational resource from @DeepMind combines GNN theory & hands-Collab explainer & coding tutorials (JAX & Jraph). It starts off gently, but does go through several core maths concepts quite in-depth.
5. Next, this follow-up @distillpub article is less introductory that resource (1), but I still find it quite accessible. It intuitively extends the convolution concept to graphs & thoroughly explains *interactively* different types of graph convolutions. distill.pub/2021/understan…
6. If papers are your thing, this 2020 paper is a good overview. I learned a lot from it.
But don't forget: #GraphNeuralNetworks are a rapidly changing field, so being on top of recent advances is a must.
7. Now, this paper is literally the most recent GNN in biomedicine review that you could possibly find, as it came up yesterday. Is is from @marinkazitnik's lab & it also discusses lots of #genomics, s.a. single cell transcriptomics or cell interactions
If you are interested in applications of GNN to genomics or 🔥biology unraveled by #singlecell#spatial#transcriptomics, I just wrote an explainer thread about two very nice recent papers that came out last week
I need to take a small detour and just mention this. I first saw Marinka's work at the ISMB conference in 2019 in Basel, where she gave a tutorial on Learning on Graphs. Her presentation was excellent & impressed first and foremost by clarity. She made it all sound so easy.
9. Next, this is a collection of hands-on coding GNN tutorials @harvardmed to which @marinkazitnik also contributed: Machine Learning on Graphs. It's quite good for coding practice.
10. And... I kept (one of) the best for last: @jure's excellent Machine Learning on Graph Stanford lectures are available on Youtube. This is a very comprehensive resource: the material covers the basics in-depth, while also diving deep towards the end.
scRNAseq cell type annotation is notoriously messy. Despite so many algorithms, most researchers still rely on manual annotations using marker genes
In a new preprint accepted at ICML GenAI Bio Workshop, we ask if reasoning LLMs (DeepSeek-R1) can help with cell type annotation🧵
Driven by @samwang36 & @RunziTan97745 & with @BoWang87, we benchmarked DeepSeek-R1-0528 on zero-shot scRNAseq cell type annotation against non-reasoning LLMs, classifiers & foundation models . What we found surprised us👇biorxiv.org/content/10.110…
Our reasoning (!) for looking into scRNAseq cell type annotation was the observation that it is a very dynamic process: despite access to so many algorithms, biomedical groups still annotate scRNAseq data manually, in an iterative process of knowledge retrieval & data assessment.
Impressive advancement in Computational Pathology.
A new multimodal foundation model by @AI4Pathology trained on 47,000 paired histology & genomics, which beautifully shows the multi-modal power of images & DNA & RNA
Even though patient genomic data is rare, it's so powerful 🧵
First, why is this model so important?
To my view, THREADS is the closest we have today to a cancer-level patient-centric foundation model.
It beautifully integrates lots of images, DNA & RNA - 3 data modalities providing critical orthogonal information about cancerous tissues
For some background:
Computational Pathology has been really revolutionized by Deep Learning (arguably like no other cancer-related field).
It turns out that the usual slides that pathologists read to diagnose & investigate tumors are very "learnable"
Many people wonder what is the scientific evidence behind what @sama & Larry Ellison said today
at The White House: that AI will cure cancer.
Truth is that this is not a hype. The potential of AI to accelerate cancer discoveries like never before is enormous.
Here’s why🧵
To start with: cancer is a very difficult problem. Funded with several billion dollars from the US government alone over the past few years, cancer survival has only marginally improved & incidence is actually increasing in younger people. That’s not good, in fact it’s really bad
We're seeing a fundamental shift in who gets cancer, moving from a predominantly male, elderly disease to one that increasingly affects women and younger people.
Key highlights 🧵
1. New Cancer Cases and Deaths in 2025:
An estimated 2,041,910 new cancer cases will be diagnosed.
Approximately 618,120 cancer deaths will occur.
This equals about 5,600 new cases and 1,700 deaths per day.
2. Progress:
Cancer death rates have declined continuously through 2022, preventing nearly 4.5 million deaths since 1991.
This progress is attributed to reduced smoking, earlier detection, and improved treatments.
New: the monthly roller-coaster through October’s coolest life science papers is here 🚀🧬
3-sentence summaries of papers on evolution, single cell methodologies, genetic screens & more.
And, only for October, an educational video on fighting cancer🤺 as a bonus.
Enjoy 3x10!
1. Assembly theory (Sharma et al., Nature)
The most (in)famous paper I read this month proposes a new framework (assembly theory, that is) to explain basically everything… or, more specifically, “to unify descriptions of evolutionary selection across physics and biology” 1/3
This paper is not an easy read for anybody (in particular evolutionary biologists), but, to its merit, it sparked scientific discussions by being different than what is expected for a scientific paper describing evolution. 2/3