Geometric Relevance Is Not Sufficient: Evaluating Focused Post-Training Supervision in Diffusion Docking
Abstract
Diffusion-based molecular docking requires learning a corrective score field across a broad corruption distribution, yet individual pose errors can differ substantially in their structural consequences. This creates a post-training conflict between preserving broad score-field coverage and concentrating supervision on corrections that appear more relevant to docking geometry. Using DiffDock, a fixed-budget study compares broad continuation with torsional supervision focused on Cartesian consequence or conditional target consistency, and with a broad/focused mixture. Evaluation separates fixed score fidelity, generated-pose quality, best-of-10 sampling capacity, and confidence-selected Top-1 docking performance. At <5 Angstrom, broad continuation had the highest mean Top-1 success among trained policies, and focused or mixed supervision showed no consistent advantage under the tested budget. The starting model had better fixed score fidelity, whereas the evaluated broad-continuation model had higher mean Top-1 success with the same aggregate best-of-10 recovery. Focusing supervision on structurally consequential or more consistently specified corrections did not consistently improve returned-pose success over broad continuation.
Related
- Molecular diffusion and binding kinetics are sufficient to explain condensate-like nuclear clustering
- A Single TCF Transcription Factor, Regardless of Its Activation Capacity, Is Sufficient for Effective Trilineage Differentiation of ESCs
- RAGCell: retrieval-augmented generation as supervision for versatile single-cell analysis
- Regulatory mechanisms of programmed cell death in hyperuricaemia and their disease relevance
- Reticulocalbin-2 (RCN2) as a Context-Dependent CREC-Family Modulator: Gene Expression and Regulation, Molecular Biology, Disease Relevance, and Translational Challenges.
BibTeX
@misc{brown2026geometric,
title = {{Geometric Relevance Is Not Sufficient: Evaluating Focused Post-Training Supervision in Diffusion Docking}},
author = {Michael Brown and Maryam Rahimimovassagh and Mehdi Yazdani-Jahromi and Sanjeda Sara Jennifer and Ivan Garibay},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.08249},
eprint = {2611.08249},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.08249},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Geometric Relevance Is Not Sufficient: Evaluating Focused Post-Training Supervision in Diffusion Docking
AU - Michael Brown
AU - Maryam Rahimimovassagh
AU - Mehdi Yazdani-Jahromi
AU - Sanjeda Sara Jennifer
AU - Ivan Garibay
PY - 2026
DA - 2026/11/05
PB - arXiv
DO - 10.48550/arxiv.2611.08249
UR - https://arxiv.org/abs/2611.08249
AB - Diffusion-based molecular docking requires learning a corrective score field across a broad corruption distribution, yet individual pose errors can differ substantially in their structural consequences. This creates a post-training conflict between preserving broad score-field coverage and concentrating supervision on corrections that appear more relevant to docking geometry. Using DiffDock, a fixed-budget study compares broad continuation with torsional supervision focused on Cartesian consequence or conditional target consistency, and with a broad/focused mixture. Evaluation separates fixed score fidelity, generated-pose quality, best-of-10 sampling capacity, and confidence-selected Top-1 docking performance. At <5 Angstrom, broad continuation had the highest mean Top-1 success among trained policies, and focused or mixed supervision showed no consistent advantage under the tested budget. The starting model had better fixed score fidelity, whereas the evaluated broad-continuation model had higher mean Top-1 success with the same aggregate best-of-10 recovery. Focusing supervision on structurally consequential or more consistently specified corrections did not consistently improve returned-pose success over broad continuation.
ER -