GANg: A Target‐Conditioned Generative Adversarial Network for Node Synthesis in Metal–Organic Framework Similarity Graphs

GANg is a target-conditioned generative adversarial network for node synthesis in metal–organic framework (MOF) similarity graphs: a Wasserstein GAN with gradient penalty (WGAN-GP) generates 106-dimensional MOF descriptors that are inserted into QMOF similarity graphs by cosine k-nearest-neighbor attachment, so graph topology is assigned rather than learned. Under matched controls and a strict inductive protocol, a conditional WGAN-GP baseline emitting the full descriptor passes every range screen while concentrating its samples and showing no conditional response (window hit rate 0.201, chance 0.200). GANg avoids these failures by emitting only an independent descriptor core—a sparsemax composition and six cell scalars—from which all derived descriptors are recomputed; it also uses a full-space critic with coverage-based checkpoint selection. It raises coverage of held-out MOFs from 0.14 to 0.81, makes every sample descriptor-schema consistent, and doubles the conditional hit rate (0.42). Pseudo-label augmentation yields modest gains only when labels are scarce (GraphSAGE −0.032 eV at 500 MOFs), is matched by simpler generators, and is driven mainly by pseudo-label distillation. GANg provides an auditable framework for evaluating synthetic graph-node hypotheses, not physically validated materials.

Authors

Publication Details

Journal
KITopen
Published
2026-11-18
DOI
https://doi.org/10.5445/ir/1000199247
Primary Topic
Machine Learning in Materials Science
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

GANg: A Target‐Conditioned Generative Adversarial Network for Node Synthesis in Metal–Organic Framework Similarity Graphs

Markus Reischl, André Orth, Mehrdad Jalali
3 links
KITopen
Machine Learning in Materials Science
article

GANg: A Target‐Conditioned Generative Adversarial Network for Node Synthesis in Metal–Organic Framework Similarity Graphs

Markus Reischl, André Orth, Mehrdad Jalali
article en

Abstract

GANg is a target-conditioned generative adversarial network for node synthesis in metal–organic framework (MOF) similarity graphs: a Wasserstein GAN with gradient penalty (WGAN-GP) generates 106-dimensional MOF descriptors that are inserted into QMOF similarity graphs by cosine k-nearest-neighbor attachment, so graph topology is assigned rather than learned. Under matched controls and a strict inductive protocol, a conditional WGAN-GP baseline emitting the full descriptor passes every range screen while concentrating its samples and showing no conditional response (window hit rate 0.201, chance 0.200). GANg avoids these failures by emitting only an independent descriptor core—a sparsemax composition and six cell scalars—from which all derived descriptors are recomputed; it also uses a full-space critic with coverage-based checkpoint selection. It raises coverage of held-out MOFs from 0.14 to 0.81, makes every sample descriptor-schema consistent, and doubles the conditional hit rate (0.42). Pseudo-label augmentation yields modest gains only when labels are scarce (GraphSAGE −0.032 eV at 500 MOFs), is matched by simpler generators, and is driven mainly by pseudo-label distillation. GANg provides an auditable framework for evaluating synthetic graph-node hypotheses, not physically validated materials.

KITopen
Machine Learning in Materials Science
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon