April Update: Building an AI project, Consulting for Microsoft, and Building a Longevity Moonshot
A quick recap of what I did in April
Hello everyone! Welcome back and I’m excited to say I have a pretty good update for April. After working in the lab last month, I realized how much potential there is for the application of AI in biotech. Its help will really accelerate discovery and innovation in the field and I’m excited to be building inside of it. I got to consult with Microsoft and start building a longevity moonshot so stick around to learn more!
I’ll try something new this time… a quote of the month! This month’s quote is:
“Live in the future, then build what’s missing.” -Paul Graham
Predicting TCR-Epitope Interactions
One of the biggest issues in immunotherapy is knowing how to find the right receptor to bind to the cancer epitope. We have molecular docking software that can predict an interaction given a protein’s structure, but this is complex and requires that you know the structure (AlphaFold can help you now). But wouldn’t it be better to just paste in two sequences and see if they interact?
That’s what I set out to do: predict if an interaction exists between a T cell receptor and an epitope sequence. Here’s a quick summary of how I did this:
Got data from the VDJdb for positive interactions of TCR and epitope sequences
Embedded those sequences using the ProtTrans T5-encoder model
Implemented and fitted an isolation forest model trained on the data
I had to use an isolation forest model as there is a very low amount of data on negative interactions (TCRs and epitopes that don’t interact). This model is able to use a method called anomaly detection which only takes the positive data and predicts if given data is an inlier or outlier. Right now, my model is at a 70% accuracy but I’m hoping to make it better with more data! Click here to read the full article!
Consulting for Microsoft
This month, I finished my consultation for Microsoft and was able to complete the deck! My team and I worked hard to find a solution to one of their biggest problems: optimizing energy usage in data centres. With the boom in AI, data centres are using a lot of energy and water, making it economically and environmentally unsustainably for Microsoft. After doing a thorough analysis, my team and I decided that a good place to save money and energy would be the cooling process. With cooling taking up 40% of the energy usage in data centres, it was crucial to lower the cost of this while also making sure the chips don’t overheat. Our solution needed to have economic incentive while also being reasonable to implement at the present moment.
Our team came up with a solution that we think can bring down energy costs for cooling down to less than 10% of the original costs. Even better, this solution doesn’t use some high-tech methods, but almost the exact opposite: using ancient architecture to have self-cooling buildings. Based on badgirs from the Middle East, this solution uses passive cooling. This process takes advantage of the temperature differentials and prevailing wind currents to take hot air out of the building while bringing cold air in. Using our solution, Microsoft will be able to save $331.9 million dollars every year.
Huge shoutout to my teammates David, Jerry, and Nico for their work on this project. We’ve been selected as a finalist so if we get chosen to pitch, I’ll let you guys know how it goes soon!
Creating a Longevity Moonshot
This project is still in the works but TL;DR I’m building a longevity moonshot with my teammate to kill senescent cells using CAR-NK therapy. Our solution uses the safety, potency, and cost-effectiveness of NK cells and applies our knowledge from cancer therapy into the longevity space.
If you want to learn more, keep posted for next month’s newsletter where I will talk more in depth about it.
Some Interesting Content
Dr Peter Attia: Scientifically Proven Ways To Build Muscle & Boost Longevity
Hopes rise for mRNA cancer vaccine after Moderna trial shows promise
That’s all for this month, stay tuned for next month! I’ll be attending RISE finalist day and also completing the longevity moonshot.