Matthew Bernstein
@matthew-bernstein.bsky.social
Principal Scientist @ Somite AI | Computational biology and biomedical data science | Boston, MA | https://mbernste.github.io
Ive seen the analogy that AI is to biological intelligence as airplanes are to birds. I think that’s appropriate. And furthermore, we find that when we make better airplanes they don’t become more bird-like. We are optimizing for something that biology didn’t optimize for. I think AI will be similar
Just as the field of ML has become a distinct discipline from statistics (but have a lot of overlap… it’s a continuum), I feel like there’s a new field of AI that is diverging from ML. Again they have a lot of overlap, but the study and use of these massive AI models feels like a new thing
It seems that computational work is undergoing two simultaneous transitions that resemble transitions of the past: 1) The transition from coding in a programming language to describing via natural language (AI writes the code) resembles the transition from machine code to programming languages.
When trying to reason or prove something in a rigorous mathematical way as it relates to a real-world problem, most of the work seems to lie in developing the right formalism. Once the formalism is established, the math becomes "disinhibited" and flows more effortlessly.
Check out my recent talk on two tools I helped develop over the last few years (SpatialCorr and Monkeybread) for analyzing spatial transcriptomics data youtu.be/R8pEvndFHyo?...
From Data to Discovery: Computational Tools for the Analysis of Spatial Genomics Data
YouTube video by Watershed Bio
youtu.be
Amazing (Here's the open access article: pmc.ncbi.nlm.nih.gov/articles/PMC...)
Science Magazine@science.org · last yr.In Science, researchers detail a nanoscale-resolution reconstruction of a millimeter-scale fragment of human cerebral cortex, giving an unprecedented view into the structural organization of brain tissue at the supracellular, cellular, and subcellular levels. scim.ag/3FvpAKy #BrainAwarenessWeek
A nice interface to track the status of various NIH websites and resources given the recent instability: stats.uptimerobot.com/Zrqh8AhvKn
Government Biomedical data
stats.uptimerobot.com
Thrilled to share Stellaromics has raised $80M in Series B to advance 3D spatial biology & launch Pyxa! This groundbreaking platform enables unprecedented visualization & analysis of cells & molecules in thick tissue samples. bit.ly/3QaGiRN #Stellaromics #Pyxa #SpatialBiology
Really interesting thesis: that the decline of American democracy and prosperity is driven largely by our declining mobility. www.theatlantic.com/magazine/arc...
How Progressives Froze the American Dream
The U.S. was once the world’s most geographically mobile society. Now we’re stuck in place—and that’s a very big problem.
theatlantic.com
Check out my recent talk on dimensionality reduction, where I try to lay out my thoughts on the topic: youtu.be/AuJzMnH78wM?...
Controversies & Challenges in Dimensional reduction and Vizualization in Single Cell Experiments
YouTube video by PANORAMICS - A Vision
youtu.be
Glad to see this issue of cell segmentation errors affecting spatial transcriptomics analysis (which most users observe) is now being investigated thoroughly: www.biorxiv.org/content/10.1...
Impact of Segmentation Errors in Analysis of Spatial Transcriptomics Data
Spatial transcriptomics aims to elucidate cell coordination within biological tissues by linking the state of the cell with its local tissue microenvironment. Imaging-based assays are particularly pro...
biorxiv.org
Here's a blog post on Euler's number: a constant I only ever partially understood despite it appearing in equations across all of math and science: mbernste.github.io/posts/eulers...
Demystifying Euler’s number
Euler’s number $e := 2.71828\dots$ has, to me, always been a semi-mysterious number. While I understood many facts about $e$, I never felt I ever truly understood what it really was – it’s core essenc...
mbernste.github.io
More of an art project than science project, The Pink Chicken project is thought provoking nonetheless pinkchickenproject.com#what
Pink Chicken Project
Pink Chicken Project suggests changing the colour of the entire species Gallus gallus domesticus to pink - using the recently invented CRISPR gene-drive technique.
pinkchickenproject.com
Cells don’t interact in just 2 dimensions—so why study them that way? In 2D spatial analysis, critical interactions between cell types can be underestimated, leading to distorted biological insights. #3DSpatial #SpatialBiology #research Image from from Sui et al., bioRxiv
Cells transport proteins in membrane vesicles. These vesicles are formed by self-assembled exoskeleton. The exoskeleton scaffolds are cuboctahedrons made of proteins. Life is amazing!
Bizarre story. Genetic Analysis of a Sarcoma Accidentally Transplanted from a Patient to a Surgeon | New England Journal of Medicine www.nejm.org/doi/full/10....
Genetic Analysis of a Sarcoma Accidentally Transplanted from a Patient to a Surgeon | NEJM
Modern concepts of cancer immunology originated from the classic observations by Jensen, Loeb, Tyzzer, and Little in the early years of the 20th century of the rejection of transplanted allogeneic ...
nejm.org
Good software engineering requires understanding human cognition to reduce cognitive load. I think all the fancy paradigms are just ways to accomplish this. minds.md/zakirullin/c...
Cognitive load is what matters
There are so many buzzwords and best practices out there, but let's focus on something more fundamental. What matters is the amount of confusion developers feel when going through the code.
minds.md
As someone new to image analysis, I recently came upon this perspective that seems trivial, but is actually important: A pixel is a "point sample" and not a little square. A pixel is a data point, sampled from the world, like any other data point would be. pages.graphics.cs.wisc.edu/559-f14/2014...
What is a Pixel? (and what is a point sample)
When we talk about image-based graphics, we talk about it being a regular collection (usually a grid) of samples (or pixels). It’s time to be a little more precise about this. The term pixel is (I’m t...
pages.graphics.cs.wisc.edu
I like ideas that find hidden biological information in seeming “artifacts”. A similar idea in this vein is RNA-velocity where intronic RNA can be used to estimate change in transcription rate.
Kieran Campbell@kieranrcampbell.bsky.social · 2y agoCell type-specific interaction analysis using doublets in scRNA-seq
Excited to finally share Perceptein, a PERCEPtron made of proTEINs: www.science.org/doi/10.1126/...
A synthetic protein-level neural network in mammalian cells
Artificial neural networks provide a powerful paradigm for nonbiological information processing. To understand whether similar principles could enable computation within living cells, we combined de n...
science.org
Understanding kernel methods, like SVMs, at a deep level requires understanding a pretty complicated mathematical object called the Reproducing kernel Hilbert Space. I wrote a blog post detailing my understanding of these objects and how they relate to kernel methods: mbernste.github.io/posts/rkhs/
Reproducing kernel Hilbert spaces and the kernel trick
If you’re a practitioner of machine learning, then there is little doubt you have seen or used an algorithm that falls into the general category of kernel methods. The premier example of such methods ...
mbernste.github.io
Over 30 prominent scientists call for a ban on the creation of a "mirror cell"--a microbe made of molecules that are mirror images of their natural forms. It could cause a mind-boggling global disaster. Here's my story [gift link] 🧪https://nyti.ms/3OUCXp6
A ‘Second Tree of Life’ Could Wreak Havoc, Scientists Warn (Gift Article)
Research on so-called mirror cells, which defy fundamental properties of living organisms, should be prohibited as too dangerous, biologists said.
nyti.ms
The 100 Prisoners Problem is a really nice and compact example of how a seemingly intractable problem can have a hidden structure that can be exploited, but utterly belies human intuition. When I first read it, I couldn't believe it had a non brute force solution. en.wikipedia.org/wiki/100_pri...
100 prisoners problem - Wikipedia
en.wikipedia.org
I think a key quality that demonstrates scientific maturity is appropriately balancing quantitative assessments with qualitative observations. Beginners tend to err too far on one side or the other: www.scientificamerican.com/article/what...
What This Graph of a Dinosaur Can Teach Us about Doing Better Science
“Anscombe’s quartet” and the “datasaurus dozen” demonstrate the importance of visualizing data
scientificamerican.com
Can AI improve the current state of molecular simulation? www.owlposting.com/p/can-ai-imp... in my first podcast, I spend 2 hours interviewing Corin Wagen and Ari Wagen, two brothers who are building the next generation of molecular simulation for drug discovery and material science
Can AI improve the current state of molecular simulation? (Corin & Ari Wagen, Ep #1)
2.1 hours listening time
owlposting.com
Come join our team and work on exciting computational problems to advance biomedical research! jobs.gusto.com/postings/ste...
High Performance Computing & Image Processing Scientist at Stellaromics Inc.
jobs.gusto.com
Three BioML starter packs now! Pack 1: go.bsky.app/2VWBcCd Pack 2: go.bsky.app/Bw84Hmc Pack 3: go.bsky.app/NAKYUok DM if you want to be included (or nominate people who should be!)
It's hard for me to imagine working in a field outside of biomedical research and computational biology. Few fields offer all of the following: 1. A vast, open frontier waiting to be explored 2. The promise of saving lives and improving human health 3. Fascinating computational problems
