Do Vision CIL Methods Transfer to Protein? A Diagnostic Benchmark of Frozen Protein Encoders
Abstract
Protein function annotation operates over an expanding label space, making class-incremental learning (CIL) a natural framework for incorporating newly discovered functions without full retraining. Yet existing CIL methods have been developed largely in vision and have not been systematically evaluated over frozen protein representations. We conduct a controlled benchmark of nine representative CIL methods across five sequence-, structure-, and sequence--structure-based protein encoders and two enzyme function datasets at different scales, while keeping the encoders frozen and evaluation settings consistent across representations. We find that no method transfers reliably across encoders, and that these failures persist across alternative phase constructions. Diagnostic analyses reveal that this instability arises from weak gradient-based stability signals, substantial heterogeneity in embedding geometry, and covariance assumptions whose appropriate capacity changes with both dataset scale and encoder. These findings suggest that the representation itself should be treated as a central component of continual-learning design rather than a passive feature extractor. More generally, reliable protein CIL will require encoder-aware methods that adapt their statistical assumptions and capacity to the geometry of the representation on which they operate. Code is available at https://github.com/zygao930/ProteinCIL.
Related
- Centered cross-context perturbation transfer on the public 2025 Virtual Cell Challenge benchmark
- Different, same, or arrhythmic: an empirically informed benchmark of differential rhythmicity methods
- A configuration-resolved benchmark of differential abundance analysis methods for human gut 16S rRNA microbiome data
- A systematic benchmark of bioinformatics methods for single-cell and spatial RNA-seq Nanopore long-read data
- A Benchmark of Unsupervised Circadian Phase Inference Methods for Bulk Transcriptomic Data
BibTeX
@misc{gao2026vision,
title = {{Do Vision CIL Methods Transfer to Protein? A Diagnostic Benchmark of Frozen Protein Encoders}},
author = {Ziyuan Gao},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.18879},
eprint = {2610.18879},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.18879},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Do Vision CIL Methods Transfer to Protein? A Diagnostic Benchmark of Frozen Protein Encoders
AU - Ziyuan Gao
PY - 2026
DA - 2026/10/17
PB - arXiv
DO - 10.48550/arxiv.2610.18879
UR - https://arxiv.org/abs/2610.18879
AB - Protein function annotation operates over an expanding label space, making class-incremental learning (CIL) a natural framework for incorporating newly discovered functions without full retraining. Yet existing CIL methods have been developed largely in vision and have not been systematically evaluated over frozen protein representations. We conduct a controlled benchmark of nine representative CIL methods across five sequence-, structure-, and sequence--structure-based protein encoders and two enzyme function datasets at different scales, while keeping the encoders frozen and evaluation settings consistent across representations. We find that no method transfers reliably across encoders, and that these failures persist across alternative phase constructions. Diagnostic analyses reveal that this instability arises from weak gradient-based stability signals, substantial heterogeneity in embedding geometry, and covariance assumptions whose appropriate capacity changes with both dataset scale and encoder. These findings suggest that the representation itself should be treated as a central component of continual-learning design rather than a passive feature extractor. More generally, reliable protein CIL will require encoder-aware methods that adapt their statistical assumptions and capacity to the geometry of the representation on which they operate. Code is available at https://github.com/zygao930/ProteinCIL.
ER -