iiinova — where your digital self keeps evolving

Eli Weinstein

@eliweinstein.bsky.social

699 followers 195 following 27 posts

Assistant professor of chemistry at the Technical University of Denmark (DTU). Also at Jura Bio. machine learning, statistics, chemistry, biophysics eweinstein.github.io

Posts Replies Media

Eli Weinstein @eliweinstein.bsky.social · 07/07/2026

We illustrate with a causal model for functional genomics that satisfies the symmetries of DNA. Methods such as AlphaGenome emerge as outcome models, while generative and masked DNA language models are the corresponding propensity models.

Diagram of a DNA sequence and genomic track, and a causal model relating the two.

110

Eli Weinstein @eliweinstein.bsky.social · 07/07/2026

These new geometric causal models describe causal relationships using equivariant maps, and can be used to draw causal inferences from complex scientific data, including spatial, graph, and molecular data.

Diagrams of sequence, spatial and array causal models.

100

Eli Weinstein @eliweinstein.bsky.social · 21/10/2025

Crucially, it depends on jointly modifying the experimental protocol and the training algorithm: on their own, neither modification helps.

100

Eli Weinstein @eliweinstein.bsky.social · 21/10/2025

This approach lets you focus limited measurements on the most informative datapoints, maximizing information gain without compromising reliability.

100

Eli Weinstein @eliweinstein.bsky.social · 21/10/2025

Second, modify the training algorithm: compensate for the missing negatives by incorporating the generative variational synthesis model into the objective.

100

Eli Weinstein @eliweinstein.bsky.social · 21/10/2025

To test, we can deliver billions of designs to different cells. But there is a cost to recovering those designs' function, to obtain (x,y) data.

100

Eli Weinstein @eliweinstein.bsky.social · 21/10/2025

With variational synthesis, we can now build quadrillions of generative model-designed sequences. The bottleneck is now testing, not synthesis.

100

Eli Weinstein @eliweinstein.bsky.social · 21/10/2025

We're excited to present LeaVS, a method to scale up learning for protein function models. It is based on the co-design of wet lab experiments and in silico training.

1114