Error-Controlled Discovery of Protein Interactions via Conformal Calibration

Abstract

Protein language models (PLMs) are powerful predictors of protein interactions, but strong benchmark performance does not ensure reliable discoveries. We introduce PPI-Caliper, a model-agnostic framework that adapts conformal calibration to select interactions at user-specified false discovery rate (FDR) levels using held-out labels without retraining the predictor. Across five species, two PLM-based predictors, and two human training-data versions for each predictor, PPI-Caliper keeps empirical FDR at or below the target level under in-distribution calibration. Under distribution shift, transferred thresholds tend to become conservative when predictive performance improves, but can fail to control FDR when performance deteriorates. PPI-Caliper connects pre-trained interaction scores to explicit error targets and demonstrates why reliable biological discovery requires evaluating calibration alongside predictive accuracy.

GoCo: structure-guided conformal calibration for error-controlled Gene Ontology annotation release
Chong Wang, Yongzhao Shao, Peng Liu
arXiv1 Oct 2026PreprintarXiv licenceFull textBioinformatics & computational biology StatisticsarXiv
Discovering host-viral protein interactions in autophagy: A LIR discovery pipeline for identifying LC3-interacting region motifs in highly virulent viruses
Kaylee D. Petraccione, Haytham M. Wahba, Mohamed G. H. Ali +8
PLOS Pathogens11 Sep 2026Peer-reviewedCC BY 4.0Full textMolecular biology & gene regulation Cell biologyPLOS
Reliability-Centered Evaluation of Sparse Longitudinal CT Lesion-Size Forecasting with Conformal Interval Calibration and Gompertz-Inspired Regularization
Lingfei Kong
arXiv18 Sep 2026PreprintarXiv licenceFull textBioinformatics & computational biology StatisticsarXiv
FiberPro 1.0: Multiagent AI-Guided Design for High-Throughput Production and Conformal Deposition of Functional Protein Micro/Nanofibers
L. Li, C. Moses, Y. Noh +6
bioRxiv18 Sep 2026PreprintCC BYFull textSystems & synthetic biologybioRxiv
Representation-aware calibration enables statistically reliable protein structure search
L. Shi, H. Zhou, Y. Yang +1
bioRxiv3 Nov 2026PreprintCC BYFull textBioinformatics & computational biologybioRxiv

Cite

BibTeX RIS

BibTeX

@misc{yang2026errorcontrolled,
  title = {{Error-Controlled Discovery of Protein Interactions via Conformal Calibration}},
  author = {Y. Yang and D. Li and D. Ding and L. Shi and H. Zhou and Y. Y. Lu},
  howpublished = {bioRxiv},
  year = {2026},
  month = nov,
  doi = {10.64898/2026.11.03.764142},
  url = {https://www.biorxiv.org/content/10.64898/2026.11.03.764142v1},
  note = {CC BY},
}

RIS

TY  - UNPB
TI  - Error-Controlled Discovery of Protein Interactions via Conformal Calibration
AU  - Y. Yang
AU  - D. Li
AU  - D. Ding
AU  - L. Shi
AU  - H. Zhou
AU  - Y. Y. Lu
PY  - 2026
DA  - 2026/11/03
PB  - bioRxiv
DO  - 10.64898/2026.11.03.764142
UR  - https://www.biorxiv.org/content/10.64898/2026.11.03.764142v1
AB  - Protein language models (PLMs) are powerful predictors of protein interactions, but strong benchmark performance does not ensure reliable discoveries. We introduce PPI-Caliper, a model-agnostic framework that adapts conformal calibration to select interactions at user-specified false discovery rate (FDR) levels using held-out labels without retraining the predictor. Across five species, two PLM-based predictors, and two human training-data versions for each predictor, PPI-Caliper keeps empirical FDR at or below the target level under in-distribution calibration. Under distribution shift, transferred thresholds tend to become conservative when predictive performance improves, but can fail to control FDR when performance deteriorates. PPI-Caliper connects pre-trained interaction scores to explicit error targets and demonstrates why reliable biological discovery requires evaluating calibration alongside predictive accuracy.
ER  -

Abstract

Related

Cite