Romain Lopez
@_romain_lopez_
Followers
2K
Following
164
Media
15
Statuses
219
Assistant Professor @ NYU Courant Institute & Biology
New York, NY
Joined March 2018
1/ I’m excited to share recent work on inferring cell-cell interactions using attention by @justjhong and me, supervised by @elhamazizi. Open the thread 🧵for a brief tweetorial of our method. bioRxiv link: https://t.co/F0YM3TYmeP
2
30
191
A simple approach for learning representations across multiple environments! Read @taromakino's work from his internship at Genentech!
When we apply machine learning to data from multiple environments (e.g. hospitals), we often learn spurious features due to the variation across environments. Our new work offers a simple solution using Supervised Contrastive Learning. https://t.co/UjYUtufHVw 1/7
0
0
13
A recording of my PhD defense is now available! https://t.co/tff79g0lT8 My thesis will be available online soon :)
9
18
156
Honored to be selected as a 2024 #STATWunderkind! Grateful for all my mentors & collaborators from @genentech and @StanfordMed. Learn more about our work here: https://t.co/Pzbyp8eiat
11
9
84
Congratulations to @JayoungR on winning the Best Paper Award at the ICML workshop on @AI_for_Science for her internship work with us at Genentech! We will also be presenting this work next month as part of an oral presentation at MLCB in Vancouver!
How can we integrate two single-cell perturbation screens like Perturb-seq and optical pooled screens? Thrilled to introduce Perturb-OT for cross-modality matching and prediction of perturbation responses! Work with amazing @_romain_lopez_ @_bunnech & Aviv (1/)
1
9
50
🚀 My close friend and former Berkeley roommate @geoffnegiar has co-founded @forecastingco with support from @ycombinator! Geoff is not only brilliant, but also kind, honest and business-savvy. Collaborators, investors, and customers will be lucky to have you! Congratulations!
Rare announcement: after finishing my PhD @berkeley_ai in 2023, and working on forecasting problems @amazon and @technology, I am launching @forecastingco with my dear friend @jodles89! We are building foundation models for time series. Get in touch!
1
2
9
Thank you so much to all of you for your kind messages and support 😀
0
0
4
This year I will start hiring at the level of postdoc and graduate student 🎓. If you're preparing graduate school applications in CS 💻 or in Computational Biology 🧬, consider applying to NYU to come work with me! More info on the group's website 🌐:
biologicalml.org
The NYU Biological Machine Learning Group led by Romain Lopez focuses on advancing machine learning tools for biological research, particularly in areas such as causality, interpretability, and...
3
12
64
Although I can’t tag everybody, I would like to give a shout out to a few people 📣: @JiaLi52524397, @KleinLabHMS, @YosefLab, Mike Jordan, Jeff Regier, @adamgayoso, @pierreboyeau, @IdoAmitLab, @jkpritch, Aviv Regev, Jan-Christian Hütter, and Kelvin Chen 🙌.
1
0
8
I am incredibly grateful for all the scientists who have helped me throughout the years 🙏. Math teachers 📐, research advisors 🧑🔬, collaborators 🤝, peer researchers 🎓, and interns 💼: it takes a village to shape someone’s research career (and likely a whole life) 🌍❤️.
1
0
9
Happy to announce that in September 2025, I will open my research laboratory at @NYU_Courant 🏢🔬. As an assistant professor of CS & Biology 💻🧬, I will carry on my work on advancing ML research for molecular biology, and apply those methods for scientific discoveries 🌟🔍.
56
31
479
How can we integrate two single-cell perturbation screens like Perturb-seq and optical pooled screens? Thrilled to introduce Perturb-OT for cross-modality matching and prediction of perturbation responses! Work with amazing @_romain_lopez_ @_bunnech & Aviv (1/)
3
31
167
Excited to present perturb-seq-in-the-loop in #RECOMB2024 next Tuesday at 2pm!! Looking forward to being back in Boston!
1/🧵Introducing Perturb-seq-in-the-loop: a sequential experimental design strategy for perturbation screens guided by multimodal priors, with 3X speedup over state-of-the-art active learning methods! With amazing @_romain_lopez_ @jchuetter Taka Kudo @antonio_science Aviv Regev
1
6
71
Finally got to celebrate with the #Starfysh masterminds (and mini-golf champions😄) @SiyuHe7 @YinuoJin6 @AchilleNazaret and all our supportive lab members ⭐️ (Yes, the paper tastes as good! 😄)
1/ #Starfysh ⭐️🐟 for spatial transcriptomic+histology analysis and multi-sample integration is out today in @NatureBiotech🎉 Thanks to reviewer feedback we improved model interpretability (learning anchors and joint representation of multiple samples) https://t.co/Hy458zzLwX
5
5
77
6/ In addition to my co-authors, a huge thank you to @seblachap for showing me basic proof techniques for non-linear ICA theory during CLeaR 2023, which helped me a great deal in getting started with this project!
1
0
7
5/ Explore our study to see how our findings could influence the development and understanding of comparative LVMs across various applications. I will present the paper at CLeaR 2024 in a few weeks!
arxiv.org
Deep Generative Models (DGMs) are versatile tools for learning data representations while adequately incorporating domain knowledge such as the specification of conditional probability...
1
0
2
4/ Here, we propose such a novel theory of identifiability for comparative deep generative models, based on non-linear ICA theory. As may be expected, the identifiability of the latent variables (by block) depends on the function class considered for the mixing function.
1
0
0
3/ Their empirical success is actually surprising, given important challenges in learning disentangled representations with LVMs. A proper identifiability theory is crucial for ensuring that the models learn meaningful and interpretable representations, but does not exist.
1
0
0