Evan Kiefl@evankiefl

Protein Bioinformatician at Arcadia Science. My opinions are my own.

Joined March 2017
11 Photos and videos
Evan Kiefl retweeted
James Golden@James_R_Golden
May 19
Glass Box UMAP computes exact feature contributions for nonlinear dimensionality reduction, and we created a Python package to make it easy to use. By @evankiefl @ryanayork @ArcadiaScience [1/n]
1
8
33
2,975
Evan Kiefl retweeted
seemay chou
@seemaychou
Feb 5
Some thoughts @ArcadiaScience on a more Bayesian approach to bio data. Everyone seems to be running towards scale/speed, but I really think we're about to hit a hard wall of diminishing returns unless we reimagine our approaches and technologies for the next era. "...the future of biological discovery won't belong to those who merely accumulate the largest haystacks or automate its brute force collection, but to those who develop the sharpest magnets for finding needles" thestacks.org/publications/i…

Biology needs to become prospective

Since biological data are often non-independent, more data doesn't always mean more insight. We argue that a prospective approach is needed to uncover the deepest principles of life.

thestacks.org
1
22
128
10,384
1/6 We published a deep dive into what MSA Pairformer learns about evolutionary relationships. @yoakiyama et al. achieved impressive contact and variant fitness prediction with a fraction of ESM2's parameters; we wanted to understand the mechanism better. tinyurl.com/4ccn397z
3
12
35
3,839
more replies
6/6 As the field shifts toward MSA-based architectures, the quality of input alignments becomes paramount. MSA Pairformer encodes evolutionary signal well, but future architectures might benefit from learning when to trust that signal.
1
174
7/6 As always, this study is end-to-end reproducible (arcadia-science.github.io/20…)
1
161
Evan Kiefl retweeted
tbepler@tbepler1
20 Jun 2025
Why does no one in AI protein engineering work on indels? We’re solving this at OpenProtein.AI. Check out our upcoming indel design tooI! 🤩 1/4
2
11
67
6,342
Our team at @ArcadiaScience took a recent preprint on attention mechanisms for genotype–phenotype mapping, validated the results, improved the performance, and turned it into shareable research in just 9 days. 1/6 🧵 arcadia-science.github.io/20…
3
23
73
20,832
more replies
5/6 This rapid iteration cycle (preprint -> validated extension in 9 days) illustrates how open science and modern publishing accelerate discovery orders of magnitude faster than traditional journals.
1
8
723
6/6 As always, code, configs, and model checkpoints are available in our repository. We hope this sparks another round of improvements! (arcadia-science.github.io/20…)
4
730
1/4 🧵 Excited to introduce "notebook pubs" — a new publication format @ArcadiaScience is experimenting with that bridges the gap between computational analysis and scientific publication! research.arcadiascience.com/…

Closing the divide between analysis and publication: The notebook pub

We're experimenting with treating our computational notebooks as publications themselves. This approach reduces publication burden, encourages faster publishing, and builds in reproducibility....

thestacks.org
2
8
20
12,793
more replies
3/4 Benefits: full reproducibility (from raw data to final output), faster sharing (publish "fresh off the keyboard"), and direct access to the scientist's working process.
1
4
308
4/4 Check out our first two notebook pubs and give us feedback! arcadia-science.github.io/20… arcadia-science.github.io/20…
6
276
Load more