Research map / Label-efficient and robust learning / Self-supervised and contrastive
Intervention-Aware Multiscale Representation Learning from Imaging Phenomics and Perturbation Transcriptomics
Jiayuan Chen, Ruoqi Liu, Zishan Gu, Ping Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add transcriptome-guided intervention-aware distillation (TIDE) on top of DINO | DINO ViT · CPG-12 one-shot unseen compounds | Top-1 accuracy (%): 36.36 -> 42.09 | Improved |
| add TIDE to DINO | DINO ViT-Small · CPG-Pilot one-shot unseen genetic perturbations | Top-1 accuracy (%): 11.36 -> 16.01 | Improved |
| add TIDE to DINO | DINO ViT-Small · RxRx3 drug-target discovery | AP: 0.300 -> 0.317 | Improved |
| add TIDE to DINO with CA-ViT | DINO CA-ViT · RxRx3 drug-target discovery | AUC: 0.545 -> 0.612 | Improved |
| add drug-molecule contrastive alignment CL(D) versus DINO | DINO · one-shot transfer (CPG-12) | Top-1 accuracy (%): 36.36 -> 40.04 | Improved |
| CL(D) alignment instead of DINO for drug-target discovery | DINO · RxRx3 drug-target discovery (ViT) | AP: 0.300 -> 0.252 | No clear effect |
| use DINO instead of MAE as SSL baseline | MAE · CPG-12 one-shot unseen compounds | Top-1 accuracy (%): 4.68 -> 36.36 | Improved |
| CA-ViT backbone instead of ViT with TIDE | TIDE · CPG-Pilot one-shot unseen genetic perturbations | Top-1 accuracy (%): 16.01 -> 17.80 | Improved |
| add chemistry-aware codebook (FCFP, learnable) versus direct drug-identity classification (ablation) | TIDE · CPG-12 one-shot unseen compounds | Top-1 accuracy (%): 39.39 -> 42.09 | Improved |
| learn the codebook end-to-end instead of freezing FCFP prototypes (ablation) | TIDE · CPG-12 one-shot unseen compounds | Top-1 accuracy (%): 40.13 -> 42.09 | Improved |
| freeze FCFP codebook versus no codebook on RxRx3 (ablation) | TIDE · RxRx3 one-shot unseen genetic perturbations | Top-1 accuracy (%): 5.20 -> 5.11 | Hurt |
| choose FCFP versus GIN molecular representation for the codebook (ablation) | TIDE · CPG-12 one-shot unseen compounds | Top-5 accuracy (%): 69.03 -> 71.07 | No clear effect |
Closest papers by meaning
- A Cross Modal Knowledge Distillation & Data Augmentation Recipe for Improving Transcriptomics Representations through Morphological Features (ICML 2025 · similarity 0.88)
- ViTally Consistent: Scaling Biological Representation Learning for Cell Microscopy (ICML 2025 · similarity 0.87)
- Learning Cross-Domain Representations for Transferable Drug Perturbations on Single-Cell Transcriptional Responses (AAAI 2025 · similarity 0.87)
- Learning Identifiable Factorized Causal Representations of Cellular Responses (NeurIPS 2024 · similarity 0.87)
- Chem-PerturBridge: a harmonized compendium of small molecule perturbation transcriptomic effects (NeurIPS 2026 · similarity 0.86)
- Learning Molecular Representation in a Cell (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
