I Finally Get Her Research

In three days, I’ll be turning 40. It’s a milestone I’ve been thinking about for a while — the passage of time, the weight of change, and the things we hope to leave behind.

I thought I’d write something reflective about this milestone. Ironically, I never completed the second part of my 39th birthday post, but then something more significant emerged.

On a date night in London, Dr. V was explaining her latest research to me. I’ll be honest — it’s not something I could grasp in one go. It’s the kind of thing that requires patience, multiple passes, and a willingness to sit with the complexity. As she explained it, something shifted within me. For her, it was simply another research project, but I could see the magnitude of it. It was impactful, seminal even. It wasn’t just another project — it was something that changes the landscape. I knew then that this work needed to be part of the Newsletter.

Since that night, I’ve spent time going deeper, parsing out the technical and scientific detail to showcase not just the depth of the work, but the magnitude of its potential. I’ve written this as a surprise for her, hoping she feels seen, understood, and celebrated when she reads it. As a caveat, Dr. V is a computational scientist and her most recent work is, in fact, in astrophysics, on Quasars, but her field is pure machine learning model-building rather than biomedicine or applied physics.

Any errors in this summary are solely my own, which I have written for a lay audience that has passing familiarity with scientific and medical literature. If you are an expert in the field, please ask me for the original published manuscript, which I am happy to send through.

Dr. Vidhi Lalchand’s (Broad Institute of MIT & Harvard) recent research, to my mind, is a genuine computational advancement in cancer drug design.

Here’s why this matters.

The Challenge: Too Many Combinations, Too Little Time

Cancer is one of the most complex diseases on Earth. No single drug can stop it. To defeat it, we need combination therapies — multiple drugs working together to overwhelm cancer cells from every angle.

But here’s the challenge:

Testing every possible drug combination takes too long, costs too much, and leaves patients waiting.

Imagine this:

• 38 drugs tested on 39 cancer cell lines at multiple doses. This means that there are 703 unique drug pairs when you combine 38 drugs two at a time. Each of the 703 unique drug pairs is tested on 39 cancer cell lines at 100 unique drug concentration pairs.

• The result? Over 2.7 million possible combinations since all pairs have to be considered.

Not all combinations are equal.

• Synergy: Some drugs work better together than expected.

• Additivity: Some drugs work as expected.

• Antagonism: Some drugs increase the viability of cancer cells, making them harder to kill.

Knowing which combination is which could mean the difference between a life-saving treatment and a missed opportunity.

That’s why predicting drug interactions before testing is so critical.

The Solution: Predict Before You Test

Dr. Vidhi Lalchand and her collaborator, Dr. Leiv Rønneberg (University of Oslo), have created a groundbreaking computational model using a technique called a Permutation Invariant Multi-Output Gaussian Process, which considers the output of all trials on all cancer cell lines simultaneously and leverages information during training to predict on unseen cancer cell lines.

Here’s the clearest way I can present it:

1. Data

• 583 drug combinations tested on 39 cancer cell lines at multiple dosage pairs.

• This yields a massive dataset with 22,737 unique data points (an existing dataset published by Merck Laboratories Boston).

2. Model Training

• The mathematical model was trained to fit a continuous manifold (a Gaussian Process) to the discrete data points in high dimensional space. Once you have access to the manifold, you can predict the viability of cancer cells at any arbitrary drug concentration pair (for the drug pairs in the training data).

3. The Kicker

• However, the real kicker is that you can also generalize to previously unseen drug pairs by leveraging their similarity to drugs in the training data. The model can now predict how any two drugs will interact — even if that exact combination has never been tested.

• No need for costly, time-consuming drug trials for every possibility.

What Makes This Model So Special?

Permutation Invariance

• The model has to treat “Drug A + Drug B” the same as “Drug B + Drug A” because it’s the same drug pair and in mathematical parlance it is called “invariance.”

• This allows for faster, smarter predictions — just like how nature works.

Multi-Output Predictions

• Traditional models predict how one drug affects one cancer cell line.

• But this model predicts how two drugs interact across 39 cancer cell lines at once — like running 22,737 lab experiments in seconds.

Three Clear Outcomes

The model tells researchers if a drug combination results in being synergistic, additive or antagonistic. These clear outcomes help researchers avoid wasting time on ineffective combinations.

Instead, they focus on the most promising drug pairings and iterate from there.

Why This Breakthrough Matters

This is a potential revolution in cancer care if the research is progressed. Here’s why:

Speed

• What used to take months & millions of dollars of manual testing can now be done instantly & in silico (on a computer instead of a hugely expensive lab).

• Instead of testing millions of combinations, researchers can focus only on the most promising ones.

Clarity

• No more guessing (well more like educated guessing; there is always a danger in blindly trusting models). The model shows which combinations work best and which to avoid.

Personalized Cancer Care

• This model allows doctors to select drug combinations that match a patient’s specific cancer type by generalizing to a cancer cell line constructed from their tumor.

In Conclusion

I hope you enjoyed this piece and I’d love to hear your thoughts. My goal is to make this newsletter a vivid lens into exciting developments — both in my world and beyond. It’s three days before I turn 40. I need to start focusing on my multi-layered celebrations (phase 1, the cottage retreat starts this weekend).

If you want to connect with Dr. V, feel free to follow her on Twitter.