@jamesbowden_

Researcher, ML for scientific design. PhD @berkeley_ai, @NSF fellow! prev @Caltech.

 Berkeley, CA
james-bowden.github.io
Joined July 2020
  • Tweets17
  • Following408
  • Followers276
1Media
I'm excited to share our ICLR paper on how leveraging decomposability can enable more efficient scientific design, with @jlistgarten and @svlevine. We were motivated by protein design in particular. Check out the blog post: james-bowden.github.io/dado/.

DADO: Leveraging Discrete Function Decomposability for Scientific Design

We introduce DADO, a method that leverages discrete function decomposability to efficiently search combinatorial design spaces.

james-bowden.github.io
1
18
4
106
59,400
James Bowden retweeted
Wei Lyu
@realweilyuOct 22
(1/6) Nirva just raised $33M to bring jewelry with perspective to the world. 🎉 $25M pre-Series A on top of our $8M seed, backed by @lightspeedvp, @southpkcommons and more. CNET Best of CES 2026. 9,000 on the waitlist. Orders open this winter. First pieces ship this year.
43
14
8
155
59,685
James Bowden retweeted
Maria Lukarska@MariaLukarskaJul 24
I’m happy to share our pre-print exploring the activity and specificity trade-offs in adenine base editors. Thanks to my amazing co-authors @loltrogge, Hunter Nisonoff and all the other contributors @CynthiaTerrace @jlistgarten @dfsavage @SavageCatsOnly biorxiv.org/content/10.64898… 🧵

Activity and specificity trade-offs in adenine base editors

Adenine base editors (ABEs) are CRISPR effectors that introduce A-T to G-C transitions in the genome using a nucleotide deaminase fused to a Cas protein. ABEs have been evolved to have very high...

biorxiv.org
1
16
1
37
8,327
James Bowden retweeted
Itay Itay@ItayTalpirJun 18
AI in Biology is an interesting field, but also a completely different ballgame if you compare it to what NLP models are capable of doing Check out our preprint, in which we systematically benchmarked protein language models in engineering tasks! biorxiv.org/content/10.64898…

Simple baselines rival protein language models in mutation-dense design tasks

Computational protein design demands generally applicable models that reliably predict or generate unmeasured variants with superior functional properties. Recent studies have proposed protein...

biorxiv.org
1
2
192
James Bowden retweeted
Stephen Lu@stephenzluJun 8
Antibody LMs learn what looks antibody-like, but not how selection turns naive germline antibodies into strong binders. @aakarshv1 and I are excited to share CoSiNE, a model that learns this germline-to-mature process for variant effect prediction and antibody design. (1/8)
8
50
7
228
47,883
Come chat w/ me about decomposability @iclr_conf poster session -- Fri 315-545, session 4 in pavilion 3, #1107. Also poke me if you'd like to grab a drink or meal together :)
James Bowden@jamesbowden_Mar 21
I'm excited to share our ICLR paper on how leveraging decomposability can enable more efficient scientific design, with @jlistgarten and @svlevine. We were motivated by protein design in particular. Check out the blog post: james-bowden.github.io/dado/.

DADO: Leveraging Discrete Function Decomposability for Scientific Design

We introduce DADO, a method that leverages discrete function decomposability to efficiently search combinatorial design spaces.

james-bowden.github.io
3
222
Our review on AI for protein engineering is out now, about this too-fast-moving field full of hype and overclaim, yet one that is having a real impact on the world and can be described in a coherent manner without histrionics science.org/eprint/666XRGRCV…

Science | AAAS

science.org
7
87
1
424
36,123
New, full length ProteinGuide manuscript, now spanning interpolative, extrapolative and multi-property (extending Pareto optimality) settings; comparisons to fine-tuning; and beating 7 rounds of directed evolution with just one round of ProteinGuide. arxiv.org/abs/2505.04823

ProteinGuide: On-the-fly property guidance for protein sequence...

Sequence generative models are transforming protein engineering. However, no principled framework exists for conditioning these models on auxiliary information, such as experimental data, without...

arxiv.org
2
14
2
56
22,741
James Bowden retweeted
Kuba
@kuba_AIMar 16
AI can optimize materials 🤘 Our (@pabbeel, @svlevine, @AIatMeta) proposed transformer model 𝗖𝗹𝗶𝗾𝘂𝗲𝗙𝗹𝗼𝘄𝗺𝗲𝗿, combined with 𝗲𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻 strategies, discovers materials that optimize target properties. arxiv.org/abs/2603.06082 – at Berkeley, CA

Offline Materials Optimization with CliqueFlowmer

Recent advances in deep learning inspired neural network-based approaches to computational materials discovery (CMD). A plethora of problems in this field involve finding materials that optimize a...

arxiv.org
5
29
2
128
59,188
Our new paper on leveraging ML to characterize the fitness landscape of two proteins binding through evolution, wrt geometry, epistasis, etc. The super hard work of Hanlun Jiang, Stephan Allenspach and @jamesbowden_ on our end. science.org/doi/10.1126/scie…

Structural ontogeny of protein-protein interactions

Understanding how protein binding sites evolve interactions with other proteins could hold clues to targeting “undruggable” surfaces. We used synthetic coevolution to engineer new interactions...

science.org
2
26
103
10,022
James Bowden retweeted
Jason Yang@jsunn_y16 Jan 2025
Happy to share that our work on Active Learning-Assisted Directed Evolution is now published in @NatureComms! We show that it's an effective and broadly applicable method to accelerate protein engineering with machine learning. Paper: nature.com/articles/s41467-0…
Jason Yang@jsunn_y29 Jul 2024
Excited to share our preprint on Active Learning-Assisted Directed Evolution (ALDE)! We present a practical workflow that leverages uncertainty quantification to efficiently navigate protein fitness landscapes. 🧵(1/6) Paper: biorxiv.org/content/10.1101/… Code: github.com/jsunn-y/ALDE

Active Learning-Assisted Directed Evolution

Directed evolution (DE) is a powerful tool to optimize protein fitness for a specific application. However, DE can be inefficient when mutations exhibit non-additive, or epistatic, behavior. Here, we...

biorxiv.org
4
17
121
19,248
James Bowden retweeted
Jason Yang@jsunn_y29 Jul 2024
Excited to share our preprint on Active Learning-Assisted Directed Evolution (ALDE)! We present a practical workflow that leverages uncertainty quantification to efficiently navigate protein fitness landscapes. 🧵(1/6) Paper: biorxiv.org/content/10.1101/… Code: github.com/jsunn-y/ALDE

Active Learning-Assisted Directed Evolution

Directed evolution (DE) is a powerful tool to optimize protein fitness for a specific application. However, DE can be inefficient when mutations exhibit non-additive, or epistatic, behavior. Here, we...

biorxiv.org
3
31
6
133
44,030
James Bowden retweeted
Yuxin Chen@yuxinch26 Jul 2023
Joint work on partition-based BO, with Caltech and LLNL collaborators @yisongyue @jamesbowden_ et al., led by Fengxue -- Looking forward to chat with you at the last #ICML2023 poster session (Th 1:30pm)!
Fengxue Zhang@zhangfx7UC26 Jul 2023
I'm excited to share our recent work on Bayesian optimization (BO) with adaptive level-set estimation, which has been accepted by ICML 23. You can find the paper on arXiv [ arxiv.org/abs/2307.13371 ] and the ICML page at icml.cc/virtual/2023/poster/….

Learning Regions of Interest for Bayesian Optimization with...

We study Bayesian optimization (BO) in high-dimensional and non-stationary scenarios. Existing algorithms for such scenarios typically require extensive hyperparameter tuning, which limits their...

arxiv.org
2
14
5,017
Bayesian modeling from first principle and memes. Let's go.
15
105
24
672
245,599
phd time @berkeley_ai @jlistgarten @svlevine !! also time to start using twitter…? super excited for the next several years and genuinely so much appreciation to @yisongyue @ryan_p_adams @YueLabCaltech @denizzokt @klbouman @countability + many others who have built me up :))
4
2
1
60
11,180
Load more