Research map / Adversarial robustness and security / Certified and empirical robustness
(Be Cautious!) Bio-Foundation Models Are Not Yet Robust to Biologically Plausible Perturbations and ML Transformations
Jinhao Duan, Ruichen Zhang, Gengwei Zhang, Huaizhi Qu, Jie Peng, Sijia Liu, Tianlong Chen
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 |
|---|---|---|---|
| apply 1% graph dissimilarity (ML graph-construction change) to GearNet (ablation) | GearNet · GO (AUPRC@MICRO) | AUPRC@MICRO: 0.7 -> 0.1 | Hurt |
| apply local residue deletion perturbation (ablation) | GearNet · EC enzyme function prediction | accuracy: 79.5 -> 64.3 | Hurt |
| apply local residue deletion perturbation (ablation) | BLAST (non-FM baseline) · EC enzyme function prediction | accuracy: 76.2 -> 76.2 | No clear effect |
| vary kNN k or radius in graph construction (ablation) | GearNet · GO-BP | AUPRC@MICRO | Hurt |
| vary kNN k in graph construction (ablation) | ESM-GearNet · EC | AUPRC@MICRO | No clear effect |
| vary radius in graph construction (ablation) | ESM-GearNet · EC | AUPRC@MICRO | Hurt |
| vary number of neighbors in graph construction (ablation) | ProteinMPNN · PInvBench mpnn validation | avg. recovery rate | Hurt |
| decrease noise level in graph construction (ablation) | ProteinMPNN · PInvBench mpnn validation | avg. recovery rate | Improved |
| vary radius cutoff (spatial proximity) versus kNN (density) (ablation) | Bio-FMs (protein structure) · protein modeling | performance drop | Hurt |
| apply rename (residue mislabel) perturbation (ablation) | ESM3 · inverse folding | result | Hurt |
| apply rename (residue mislabel) perturbation (ablation) | ProNet · function prediction | result | No clear effect |
| increase coordinate Gaussian noise severity above four (ablation) | ESM3 and ProNet · biological perturbation | result | Hurt |
Closest papers by meaning
- Contextualizing biological perturbation experiments through language (ICLR 2025 · similarity 0.86)
- PertEval-scFM: Benchmarking Single-Cell Foundation Models for Perturbation Effect Prediction (ICML 2025 · similarity 0.86)
- PerturBench: Benchmarking Machine Learning Models for Cellular Perturbation Analysis (NeurIPS 2025 · similarity 0.85)
- Specialized Foundation Models Struggle to Beat Supervised Baselines (ICLR 2025 · similarity 0.85)
- PerturbReason: A Knowledge-Grounded Benchmark and Framework for Cell-State–Conditioned Mechanistic Reasoning of Perturbation Effects (NeurIPS 2026 · similarity 0.85)
- A benchmark for prediction of transcriptomic responses to chemical perturbations across cell types (NeurIPS 2024 · similarity 0.83)
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.
