Recycle only when you have to!
Abstract
AlphaFold 3-like models (e.g., Boltz, Protenix, etc.) have pushed the frontier of protein binder design. This architecture is used both for the generation of binder candidates (e.g., by hallucination or converting these models into backbone or full atom generative models) as well as for the screening of these candidates for valid binders. In this paper, we start from the observation that AlphaFold 3's confidence metrics (ipSAE, ipTM) as well as structural consistency metrics can be predicted directly from the trunk even at early recycling. Because our proposed predictor is cheap, differentiable, and available at every cycle, it can be used in two distinct ways: as a decision signal to wisely spend the recycling budget for screening and as an objective for hallucination-based design. We show that both applications lead to increased performance at significantly reduced computational cost.
Related
- Coral restoration is only practical when local human pressure levels are low
- Identification and Expression Patterns of Opsin Genes in Phortica okadai, the Only Confirmed Vector of Thelazia callipaeda in China
- A foundation model learns the sequence and functional grammar of fully human heavy-chain-only antibodies
- Orchids from Rio Grande Do Sul, Southern Brazil: What We Know and What We Still Have to Know to Ensure Their Conservation
- LGR5 and SOX2 expressing progenitor cells in the adult and aged human and mouse inner ear have the potential to produce Myosin 7A positive cells in vitro
BibTeX
@misc{vaitl2026recycle,
title = {{Recycle only when you have to!}},
author = {Lorenz Vaitl and Ivan Tanasijevic and Arian Rokkum Jamasb and L. Santiago Mille-Fragoso and Pan Kessel},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.40665},
eprint = {2610.40665},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.40665},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - Recycle only when you have to!
AU - Lorenz Vaitl
AU - Ivan Tanasijevic
AU - Arian Rokkum Jamasb
AU - L. Santiago Mille-Fragoso
AU - Pan Kessel
PY - 2026
DA - 2026/10/09
PB - arXiv
DO - 10.48550/arxiv.2610.40665
UR - https://arxiv.org/abs/2610.40665
AB - AlphaFold 3-like models (e.g., Boltz, Protenix, etc.) have pushed the frontier of protein binder design. This architecture is used both for the generation of binder candidates (e.g., by hallucination or converting these models into backbone or full atom generative models) as well as for the screening of these candidates for valid binders. In this paper, we start from the observation that AlphaFold 3's confidence metrics (ipSAE, ipTM) as well as structural consistency metrics can be predicted directly from the trunk even at early recycling. Because our proposed predictor is cheap, differentiable, and available at every cycle, it can be used in two distinct ways: as a decision signal to wisely spend the recycling budget for screening and as an objective for hallucination-based design. We show that both applications lead to increased performance at significantly reduced computational cost.
ER -