Research map / Large language models / Agents, code and math
CP-Agent: Context‑Aware Multimodal Reasoning for Cellular Morphological Profiling under Chemical Perturbations
Yuxin Zhang, Yiyao Li, Ping Shu Ho, Simon See, Zhenqin Wu, Kevin Kin-Man Tsia
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 |
|---|---|---|---|
| use CP-CLIP context-aware training instead of plain CLIP | CLIP ViT-B/16 · Unseen-drug matching | average cosine similarity: 0.286 -> 0.432 | Improved |
| use continuous descriptor drug encoding instead of binary fingerprint (ablation) | CP-CLIP ViT-B/16 · Unseen-drug matching | average cosine similarity: 0.360 -> 0.432 | Improved |
| scale vision encoder from ViT-B/16 to ViT-L/16 (ablation) | CP-CLIP (descriptor) · Unseen-drug matching | average cosine similarity: 0.432 -> 0.444 | Improved |
| use sigmoid (SigLIP) objective instead of CLIP softmax contrastive loss (ablation) | CP-CLIP (descriptor) · Unseen-drug matching | average cosine similarity: 0.432 -> 0.414 | Hurt |
| scale vision encoder from ViT-B/16 to ViT-L/16 (ablation) | CP-CLIP (descriptor) · Seen-drug compound classification | macro-avg F1: 0.896 -> 0.891 | No clear effect |
| use fingerprint instead of descriptor drug encoding (ablation) | CP-CLIP ViT-B/16 · Seen-drug compound classification | macro-avg F1: 0.896 -> 0.887 | No clear effect |
| use CP-CLIP descriptor instead of CLIP ViT-B/16 | CLIP · Seen-drug compound classification | macro-avg F1: 0.657 -> 0.896 | Improved |
| use CP-CLIP instead of general MLLMs (Gemini-2.5-Pro) | CP-CLIP ViT-B/16 (descriptor) · Seen-drug compound classification | macro-avg F1: 0.007 -> 0.896 | Improved |
| use CP-CLIP ViT-B/16 (descriptor) instead of CLIP | CLIP · Seen-drug similarity (averaged) | cosine similarity: 0.467 -> 0.549 | Improved |
| fine-tune VISTA-2D on Cell Painting data | VISTA-2D · Instance segmentation | PQ: 0.151 -> 0.682 | Improved |
| fine-tune VISTA-2D on Cell Painting data | VISTA-2D · Instance segmentation | IoU: 0.272 -> 0.824 | Improved |
Closest papers by meaning
- MorphoHELM: A Comprehensive Benchmark for Evaluating Representations for Microscopy-Based Morphology Assays (NeurIPS 2026 · similarity 0.85)
- BioXArena: Benchmarking LLM Agents on Multi-Modal Biomedical Machine Learning Tasks (NeurIPS 2026 · similarity 0.84)
- CPathAgent: An Agent-based Foundation Model for Interpretable High-Resolution Pathology Image Analysis Mimicking Pathologists' Diagnostic Logic (NeurIPS 2025 · similarity 0.84)
- How Molecules Impact Cells: Unlocking Contrastive PhenoMolecular Retrieval (NeurIPS 2024 · similarity 0.84)
- CellCLIP - Learning Perturbation Effects in Cell Painting via Text-Guided Contrastive Learning (NeurIPS 2025 · similarity 0.84)
- ScreenShot: A Foundation Model for Few-Shot Combination Drug Screening (NeurIPS 2026 · similarity 0.84)
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.
