Eli Weinstein
@EliWeinstein6
Followers
859
Following
264
Media
32
Statuses
137
Postdoc at Columbia in @blei_lab // @jura_bio // statistics, biophysics, machine learning Incoming assistant professor of chemistry at DTU in Copenhagen.
Joined May 2021
RT @lizbwood: So pleased our work on #VariationalSynthesis was selected as one of the top 4 papers at @AIHealthMIT #MoML2024. @EliWeinstein….
0
5
0
If you know of positions that might be a good fit, I'd be very grateful if you reached out.
1
1
2
I am on the academic job market this year! My research advances probabilistic machine learning and its application to biology. I'm looking for faculty positions in stats/CS/applied math and in bio departments. Academic website:
2
16
79
The paper is and the application to immunology he mentioned is now out too
0
0
2
Thanks to everyone who came to Dave @blei_lab's talk about our work on hierarchical causal models - the recording is now up:
Thank you to David Blei for kicking off our fall colloquium series -- we had a full house! . 🎥 Missed it? Stay tuned for David's recording. 📣 Curious about our next event? Go to to learn about our upcoming colloquium this Friday with Mustafa Khammash.
1
1
13
RT @Schmidt_Center: Happy Monday! Missed @Columbia's Professor of Statistics and Computer Science David Blei's colloquium from September on….
0
3
0
RT @etowah0: Getting data from human biology is the gold standard in drug development. But how do you get it without running a clinical tra….
0
2
0
If you'd like to learn more you can check out the paper plus @jura_bio 's blogpost focusing on implications for drug discovery.
0
0
5
and to other diseases, especially cancers, where immune receptor-based therapeutics are particularly important.
1
0
2
Going forward, we hope to extend this approach to B cell receptors, to inform antibody-based medicines, ….
1
0
2
And T cell vaccine candidates: epitopes that bind TCRs with strong positive effects on patient outcomes.
1
0
2
We also find potentially therapeutic TCRs: sequences estimated to cause strong positive patient outcomes.
1
0
2
We uncover complex heterogeneity of TCR effects within patient repertoires, including TCRs that cause worse patient outcomes (maybe related to hyperinflammation, etc.)
1
0
4
We demonstrate by estimating the effects of TCRs on COVID-19 severity.
1
0
2
In other words, we obtain a mapping from amino acid sequence to clinical outcomes, which we can use for therapeutic design, etc.
1
0
3
The output is a trained neural network which estimates the causal effect of adding any given TCR sequence (seen or unseen) to patient repertoires.
1
0
1
We develop an estimator for our identification formula using deep representation learning. The method scales to repertoire datasets with a hundred million TCR sequences or more.
1
0
1
We establish causal identification by using a biophysical model of T cell development as a semiparametric constraint on the causal model.
1
1
3
Since there are many TCRs within each patient, we use our framework of hierarchical causal models
1
0
2
We use the pre-selection repertoire as an instrumental variable to correct for confounding.
1
0
1