For everyone going to #SfN in San Diego

A lot of contributions here, even new ways to define authorship order

The spot of a past lab member usually gets scavenged by other members rather quickly, on a first-come, first-served basis. This time, I'm separating my most valuable possessions and making a raffle so that the other lab members have equal opportunities in the scavenging

Last week I defended my Ph.D. It's a bittersweet moment to say goodbye to a project that has shaped my life for years. Time to look toward what's next.
To everyone who's been part of this journey, thank you.
I also got this slick katana with a manifold engraved in it, from @lukesjulson.bsky.social

Lab care package I keep by my desk at all times

Am I ready to defend my thesis yet

Want to explore gcPCA?
I have prepared a detailed tutorial to help you get started:
github.com/SjulsonLab/generalized_contrastive_PCA/tree/main/tutorial
10/

We packaged everything in the gcPCA toolbox, an open-source package with multiple solutions for different needs:
📂 github.com/SjulsonLab/generalized_contrastive_PCA
- Asymmetric or symmetric, Orthogonal or non-orthogonal, and sparse solutions
👉 Check out Table 1 in the paper for details!
9/

Using the sparse solution for gcPCA, we identified multiple genes previously linked to diabetes to be co-expressed in pancreatic type II diabetes patients
8/

gcPCA allowed us to identify subtle, biologically meaningful patterns. For example, when analyzing pre- and post-learning hippocampal activity, PCA returns components with no apparent task structure. However, gcPCA revealed components with a structure reflecting the task performed.
7/

So we developed gcPCA, a flexible approach that builds on the strengths of cPCA while addressing its limitations.
The key idea?
We add a normalization in the objective function to identify dimensions with the largest relative change in variance between conditions.
6/

But it’s not as simple as just “removing” the hyperparameter. In real-world biological data, especially when sample sizes are small, random fluctuations in high-variance dimensions can overshadow the true signal.
We believe cPCA introduced the α to try to suppress these fluctuations.
5/

That’s when the idea for generalized contrastive PCA (gcPCA) was born.
What if we could:
- Remove the hyperparameter α?
- Make the method symmetric, treating both datasets equally.
4/

We initially tried using contrastive PCA (cPCA), which showed promise. However, it came with a drawback:
- A hyperparameter (α) controls the comparison, and different (α) values give equally probable solutions.
- It uses one experimental condition as a control, creating asymmetric comparisons.
3/

Our team works with high-dimensional datasets, think large-scale electrophysiology, single-cell RNA-seq, etc. In our last project, we hit a wall: how do you compare two experimental conditions (e.g., asleep vs. awake neural activity) when existing tools focus on one dataset at a time?
2/

Does your research involve comparing experimental conditions? Then our latest publication is for you: We developed generalized contrastive PCA (gcPCA), a tool for comparing high-dimensional datasets. 🧠📊 doi.org/10.1371/journal.pcbi.1012747
This tool was born out of necessity, here is the story. 🧵
1/

Doom running on a PDF made my day

github.com/ading2210/do...