@edhirata.bsky.social
🇲🇽🇯🇵 Computer Vision and Machine learning engineer at CZ Biohub SF || Ph.D || views are my own || (he/him)
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
Grateful to our incredible collaborators! This work wouldn’t have been possible without the deep interdisciplinary collaboration across CZ Biohub teams.
These fast iterations and bold ideas are only possible because of you! 🙏 #CZBiohub @czbiohub.bsky.social
#science #CellBiology #AI #microscopy
We provide a:
- PyTorch code for training & inference: github.com/mehta-lab/Vi...
- A GUI to visualize trajectories in real and embedding space
- Tools for efficient human-in-the-loop annotation
DynaCLR’s temporally consistent embeddings enable in silico synchronization of asynchronous dynamics like infection & mitosis, along with organelle responses, enabling pseudotime estimation across multiple imaging channels. 5/⏳
06.07.2025 01:15 — 👍 0 🔁 0 💬 1 📌 0
We introduce a strategy to accelerate label-free cell state analysis: cross-modal knowledge distillation transfers labels from easy-to-annotate fluorescence sensors to unlabeled phase images, unlocking multi-state measurements from label-free data. 4/🧑🏫
06.07.2025 01:15 — 👍 0 🔁 0 💬 1 📌 0DynaCLR embeddings enable efficient classification of cell states, including mitosis and viral infection, with a few thousand point annotations of cell state, which speeds up measurement of complex cell state dynamics. 3/🦠
06.07.2025 01:15 — 👍 0 🔁 0 💬 1 📌 0
DynaCLR learns embeddings from 3D multi-channel movies and enables:
1) cell state classification
2) fluorescence-to-label-free label transfer
3) alignment of dynamics.
This update adds knowledge distillation and temporal alignment for even richer analysis. 2/🔬
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
#sciencejobs #imageprocessing #bioimage #zarr #xarray @czbiohub.bsky.social
27.03.2025 04:40 — 👍 0 🔁 0 💬 0 📌 0
We're looking for a software engineer (recent grad BS/MS) excited to build and deploy high-performance tools for dynamic imaging of cells, tissues, and organs 🐟🦠🔬 — think terabytes of image data, cell phenotyping, and high-throughput microscopy.
job-boards.greenhouse.io/chanzuckerbe...
