Research map / Probabilistic and scientific modeling / Molecules and proteins
MUST-CNN: A Multilayer Shift-and-Stitch Deep Convolutional Architecture for Sequence-Based Protein Structure Prediction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| per-task fine-tuning of the multitask large model (ablation) | MUST-CNN large · 4prot dssp | Q8 accuracy: 69.5 -> 76.7 | Helped |
| per-task fine-tuning of the multitask large model (ablation) | MUST-CNN large · 4prot ssp | Q3 accuracy: 82.5 -> 89.6 | Helped |
| per-task fine-tuning of the multitask small model (ablation) | MUST-CNN small · 4prot dssp | Q8 accuracy: 67.0 -> 70.6 | Helped |
| per-task fine-tuning of the multitask small model (ablation) | MUST-CNN small · 4prot sar | Q2 accuracy: 79.0 -> 81.2 | Helped |
| scale up from small (189 hidden units) to large (1024 hidden units) model (ablation) | MUST-CNN · 4prot dssp | Q8 accuracy: 67.0 -> 69.5 | Helped |
| add multilayer shift-and-stitch with maxpooling (ablation) | MUST-CNN · 4prot | average accuracy: almost 0.5% | Helped |
| tweak hyperparameters within optimal learning rate (ablation) | MUST-CNN · 4prot | average accuracy: <1% | No clear effect |
| MUST-CNN large vs LSTM on CullPDB/CB513 | protein secondary structure · CB513 | Q8 accuracy: .674 -> .684 | Helped |
| fine-tune the large model (separate models per task) (ablation) | MUST-CNN large · 4prot test time | ms per million amino acids: 553 -> 1597 | Hurt |
Closest papers by meaning
- Clustering for Protein Representation Learning (CVPR 2024 · similarity 0.83)
- Protein Multimer Structure Prediction via Prompt Learning (ICLR 2024 · similarity 0.83)
- From Mechanistic Interpretability to Mechanistic Biology: Training, Evaluating, and Interpreting Sparse Autoencoders on Protein Language Models (ICML 2025 · similarity 0.83)
- Multi-Scale Representation Learning for Protein Fitness Prediction (NeurIPS 2024 · similarity 0.82)
- Deep Reinforcement Learning for Modelling Protein Complexes (ICLR 2024 · similarity 0.82)
- FoldToken: Learning Protein Language via Vector Quantization and Beyond (AAAI 2025 · similarity 0.82)
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.