We're inspired by work in the late 70s by Harold Scheraga using differential geometry to describe polymer conformations (pubs.acs.org/doi/epdf/10.102…) and a @MLevitt_NP2013 1983 JMB classic (sciencedirect.com/science/ar…) calculating writhe of protein conformations using discrete segments.

[8/] 📝 Paper: openreview.net/forum?id=L2jj… 🧬 Repo: github.com/Scientific-Comput… 📺 YouTube: youtu.be/Mjj6MfVQCNM?si=0FX_… With Zohar Rosenwasser, Erez Levanon, @MLevitt_NP2013

I’ve recently updated German data to use more user-friendly 10-year age groups instead of the previous 5-year intervals. The difference was surprisingly large, so I decided to investigate—digging deep into the code for about three days to verify everything. Here’s what I found: Overview of Mortality Watch Data Handling Mortality Watch integrates multiple data sources and can handle different temporal resolutions (yearly, monthly, weekly). A few months ago, I refined the logic to select the best available data for each resolution—e.g., using yearly data for yearly resolutions. For these recent updates, I also added 5-year resolution yearly data: • Yearly Deaths: These match exactly with the underlying dataset from Destatis, which provides official yearly figures. • Yearly Population: This is reported as of December 31st each year. The simplest approach would be to join the data by year, but that would be inaccurate. Since the beginning of Mortality Watch, I’ve always calculated daily population estimates—breaking down deaths and population into daily values and interpolating as needed. This ensures consistency, as summing up daily data correctly matches the totals. When comparing Mortality Watch population data with the Destatis source, Mortality Watch reports the mean of daily interpolated data. This may seem counterintuitive at first, but it provides a more accurate representation. Verifying the Accuracy of Mortality Watch To ensure my calculations were correct, I conducted a deep dive and even discovered a small bug. The issue, which had only a minor impact, affected age-stratified CMR calculations: github.com/MortalityWatch/ch… To further validate the methodology, I created an example script demonstrating different approaches to calculating ASMR. The results vary significantly based on the method used: 1️⃣ Simple year joins (joins deaths of 2020 to population of 12/31/2020), using Mortality.org age groups (0–14, 15–64, 65–74, 75–84, 85+) 2️⃣ Simple year joins, using 10-year age groups (0–10, …, 80+) – Currently used on Mortality Watch 3️⃣ Daily year joins (interpolated daily population joined on daily death data (evenly split)), using 10-year age groups (0–10, …, 80+) 4️⃣ Simple year joins, using single-year age groups (0–85) 5️⃣ Daily year joins, using single-year age groups As you can see, the red values match MortalityWatch figures identically, (besides 2023 🥴). It’s good news that there aren’t any bugs in the calculations. It’s also important to be aware of the sensitivity of the methodology used, and that we should always use the highest resolution possible. This analysis demonstrates how different age groupings and interpolation methods can have a significant impact on ASMR calculations.

New mRNA vaccines on arrival? 🤔

@GalOren8 has done something remarkable in AI for Science by teaching ChatGPT the language of RNA. I am so proud to be involved in this project with him and Erez Levanon.

OT, I encountered your sensible videos and interviews in the early days of the covid lockdowns from Australia. Just want you to know i appreciated your comments and interviews back then. Cheerio.

Okay. Thanks for the nerd snipe guys. I spent the day learning exactly how DeepSeek trained at 1/30 the price, instead of working on my pitch deck. The tl;dr to everything, according to their papers:

Really great interviews and follow-up by Roni Dori @RoniDori Actually I am Michael Levitt @MLevitt_NP2013. I will post the link English translation if there is any interest. This could my first Covid-free post since March 2020.

Yes yes, I’ll try and finish my quadruple check for bugs in the fitting algorithm asap. So much time and so little to do…

Deepseek R1 is one of the most amazing and impressive breakthroughs I’ve ever seen — and as open source, a profound gift to the world. 🤖🫡

As far as I know Micheal Levitt still supports the notion that pandemic virus spread follows a gompertz curve, like he stated in 2020.

כתבה מיוחדת עבור הקורא נתניהו, שפעל, מסיבות לא ידועות, בדיוק ההיפך מהנדרש למיטוט ארגון טרור דרך הכיסים - וריפד את כיסיו של החמאס היטב לאורך שנים.