Alishba Imran
@alishbaimran_
CS @berkeley_eecs | curr: ML bio @arcinstitute, research @berkeley_ai | prev: @czbiohub, @tesla, @NVIDIA, founded ML battery startup
The AI for Robotics e-book is out now!🎉 450 pages, 200+ visuals, 150K words covering perception, 3D sensor fusion, foundation models, transformers & diffusion for control, sim, RL & more. Available now: Nature Springer: link.springer.com/book/10.1007/9… Amazon: amazon.com/dp/B0F7CBJS52
Dropped the Virtual Cell Challenge Primer on HF. We are shipping transformers support for STATE (the SOTA model for predicting perturbation response) very soon!
The virtual cell is a longstanding vision, a tool to guide experiment design, understand function My work uses AI to build a platform for engineering cell state, a capability needed to realize this vision Come learn about our model development + AI-guided data generation 1/8🧵
Following the amazing talk today by Emma Dann we are excited to host @yusufroohani of the @arcinstitute for a special seminar this Thursday @ 10AM, presenting the efforts towards a virtual cell platform🔮
elegant approach for designing tf payloads for epigenetic reprogramming which enables in silico prediction of cell state changes from sparse combinatorial data:
reprogramming cells with transcription factors is our most expressive tool for engineering cell state traditionally, we found TFs by ~guesswork @icmlconf we're sharing @newlimit's SOTA AI models that can design reprogramming payloads by building on molecular foundation models
This review from the director of AI research at @microsoft perfectly captures why we wrote our book AI for Robotics! “… reframing classic robotics challenges through a deep learning lens” We always find reviews/feedback like this really helpful! 🙏
"The body of data available in protein sequences is something fundamentally new in biology and biochemistry, unprecedented in quantity, in concentrated information content and in conceptual simplicity." - Margaret Dayhoff describing my research better than me before I was born
Reading this book, gotta say liking it so far. Traditional robotics books (as important as they are for foundational knowledge) don't focus much on the AI part and are much more focused on control systems. Thank you @keerthanpg @alishbaimran_
We have a fun update on DynaCLR (arxiv.org/abs/2410.11281), a self-supervised method for learning cell state dynamics from 3D multi-channel movies! Congratulations, team @czbiohub, for reaching this milestone! Special thanks to Carolina Arias and her team for the close…
Excited to share an update on #DynaCLR—a self-supervised method to learn dynamic cell & organelle embeddings from time-lapse microscopy using contrastive learning. DynaCLR combines single-cell tracking & time-aware sampling for robust representations. arxiv.org/abs/2410.11281
Some of the most exciting applications of deep learning are in biology. Really fascinating work
Excited to share our new paper on DynaCLR, a self-supervised model for learning cell dynamics from terabytes of live imaging! Enables: - Detection of infection & cell division - Label-free prediction of states - Alignment of async cell trajectories - Organelle remodeling (1/4)
Excited to share an update on #DynaCLR—a self-supervised method to learn dynamic cell & organelle embeddings from time-lapse microscopy using contrastive learning. DynaCLR combines single-cell tracking & time-aware sampling for robust representations. arxiv.org/abs/2410.11281
We’re also excited about this result! Zero-shot prediction showed clear value of embeddings: - Pretraining State on Tahoe-100M - Fully fine-tuning on smaller, noisier datasets - Led to more accurate perturbation ranking prediction than mean baselines or HVG-trained State models
The result I'm most excited about from Arc's new State model: The ability to generalize on zero-shot out-of-distribution predictions after pre-training on the TAHOE-100M data set. Whereas PLMs have seemingly benefitted less from scaling data and model size, this is an inkling…
Following the launch of STATE, Arc is launching a new challenge! Build ML models to predict how human cells respond to perturbations, using our new H1 human embryonic stem cell line data. Check it out:
Announcing the inaugural Virtual Cell Challenge! Hosted by Arc Institute, and sponsored by Nvidia, 10x, and Ultima, help solve one of biology’s biggest challenges with AI by building cell state models that accurately predict responses to perturbation. virtualcellchallenge.org
Introducing Arc Institute’s first virtual cell model: STATE