Quartz 4

❯

❯

❯

Generalization

Created Sep 26, 2026Modified Sep 07, 2026

Memorization, leakage, out-of-distribution performance, shortcuts, robustness, and when apparent progress does not transfer.

49 items with this tag.

Sep 25, 2026
General-purpose language models underperform specialist models on tested epitope identification tasks
- antibodies/recognition
- evidence/generalization
Sep 25, 2026
Linear models remain competitive on epistasis-enriched protein fitness tests
- prediction/variant-effects
- evidence/generalization
Sep 07, 2026
ESM-IF, but not other inverse folding models, has learned some evolutionary constraints from sequence databases
Aug 25, 2026
Pretraining performance does not capture effectiveness on downstream tasks
Aug 25, 2026
Protein backbone design methods show fold-level novelty by recycling memorized subdomain structures
- design/backbones
- evidence/generalization
Aug 25, 2026
The Boltz-2 affinity module cannot be effectively repurposed for PPI affinity prediction
- prediction/binding
- evidence/generalization
Aug 19, 2026
Fine-tuned protein language models can struggle to generalize to positions absent from fine-tuning data
Jul 28, 2026
All-atom structure and affinity predictors partially generalize to point mutants
Jul 28, 2026
PLMs learn family-specific protein contacts from sequence context windows of about 20-40 amino acids
- model-analysis/interpretability
- evidence/generalization
Jul 28, 2026
Random splits overestimate protein language model generalization on antibody expression prediction
- evidence/generalization
- prediction/stability-expression
Jul 28, 2026
Structure prediction and design tools trained on monomers generalize to oligomers
Jul 20, 2026
MSAs make AlphaFold3 predictions robust to extensive mutation and deletion
Jul 20, 2026
Protein language models are in-context learners that default to retrieval over reasoning
- model-analysis/interpretability
- evidence/generalization
Jul 17, 2026
Language models cannot extrapolate to functional novelty or ultra-high-fitness variants
Apr 21, 2026
AlphaFold2 recapitulates interaction biases from PDB when modeling antibodies
- evidence/generalization
- prediction/complexes
Apr 21, 2026
Correct antibody-antigen prediction in AF3 and related models is partially determined by training set similarity
- evidence/generalization
- prediction/complexes
Apr 21, 2026
All-atom protein structure prediction methods are unable to model PPIs with large buried surface areas
- prediction/complexes
- evidence/generalization
Apr 21, 2026
AlphaFold2 generalizes to unseen secondary structures and topologies
- evidence/generalization
- prediction/structure
Apr 21, 2026
AlphaFold3 universally predicts the active state of ligand-bound GPCRs, even when the ligand is an antagonist
- prediction/ensembles
- evidence/generalization
Apr 21, 2026
Biophysical models trained on single and double point mutations can predict effects of three or more mutations on stability
Apr 21, 2026
Closely related protein language models have diverging performance on specific domains
- evidence/generalization
Apr 21, 2026
Composition of negative data affects rule discovery in binary classification tasks
Apr 21, 2026
DL ligand docking methods generate unrealistic poses
- prediction/ligand-docking
- evidence/generalization
Apr 21, 2026
DL structure prediction methods cannot predict CDR conformations unseen during training
- evidence/generalization
- prediction/antibody-structure
Apr 21, 2026
Deep learning methods cannot generalize T-cell receptor binding to new epitopes
- evidence/generalization
- prediction/binding
Apr 21, 2026
Inverse folding generalizes to unseen structures better than PLMs
- evidence/generalization
- design/inverse-folding
Apr 21, 2026
Language-based protein folding NNs predict novelty with far lower confidence than MSA-based protein folding NNs
- evidence/generalization
- prediction/structure
Apr 21, 2026
ML models trained exclusively on experimental structures are less effective on computational models
Apr 21, 2026
PLMs downweigh probability of sequences with multiple mutations
- prediction/variant-effects
- evidence/generalization
Apr 21, 2026
Protein folding neural networks can be subject to adversarial attacks
- evidence/generalization
- prediction/structure
Apr 21, 2026
Protein folding neural networks do not extrapolate to new ligand binding sites
- evidence/generalization
- prediction/structure
Apr 21, 2026
Protein language model embeddings are more predictive of homology than catalytic efficiency
Apr 21, 2026
Protein language models and PLM-based structure prediction generalize to de novo designed proteins
Apr 21, 2026
Protein language models are better zero-shot predictors for ranking closely related sequences than distantly related sequences
- prediction/variant-effects
- evidence/generalization
Apr 21, 2026
Protein language models can predict zero-shot which proteins belong to the same species
Apr 21, 2026
Protein structure prediction methods are unable to predict the energetics of a conformational landscape unless explicitly trained for that purpose
- prediction/ensembles
- evidence/generalization
Apr 21, 2026
Protein-ligand co-folding methods do not generalize beyond their training set
- evidence/generalization
- prediction/ligand-docking
Apr 21, 2026
Sequence-based variant effect prediction methods generalize to de novo proteins
- evidence/generalization
- prediction/variant-effects
Apr 21, 2026
Unfolded and nonfunctional isoforms of folded proteins are incorrectly predicted as folded by protein folding neural networks
- evidence/generalization
- prediction/structure
Apr 21, 2026
ddG predictions calculated from AlphaFold structures are equally accurate as those derived from experimental structures
- prediction/stability-expression
- evidence/generalization
Apr 21, 2026
Hybrid protein sequence-structure ML models rely more on sequence when many homologs were available during training
- model-design/multimodal
- evidence/generalization
Apr 21, 2026
All-atom structure prediction of RNA is driven by memorization
- evidence/generalization
- prediction/complexes
Apr 21, 2026
Biomolecular diffusion models cannot reproduce the equilibrium dynamics of the simulations they are trained on
Apr 21, 2026
De novo designed proteins (from language models) and naturally occurring proteins borrow heavily from other functionally similar sequences
- evidence/generalization
- design/sequence-generation
Apr 21, 2026
Linear MD encoders outperform nonlinear encoders when extrapolating to unseen regions of protein conformational space
- biophysics/ensemble-analysis
- evidence/generalization
Apr 21, 2026
Memorization in protein structure prediction
- evidence/generalization
- prediction/structure
Apr 21, 2026
Models trained on synthetic data tend to converge to mean
- evidence/datasets
- evidence/generalization
Apr 21, 2026
Poisoning attacks on LMs require a constant number of samples regardless of scale
- evidence/generalization
- evidence/datasets
Apr 21, 2026
Viral protein MSAs fail to recapitulate contacts from evolutionary coupling analyses
- evolution/homology
- evidence/generalization

Created with Quartz v4.5.2 © 2026

GitHub
Discord Community