# Fragment-Level Graph Encoding for Interpretable Molecular Property Prediction

DevFeed: [Fragment-Level Graph Encoding for Interpretable Molecular Property Prediction](<https://devfeed.tech/articles/fragment-level-graph-encoding-for-interpretable-molecular-property-prediction-94959.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.29589>)

Author: Rafael Lopes Almeida (School of Pharmacy, University of Eastern Finland, Kuopio, Finland), Ina P\\"ohner (School of Pharmacy, University of Eastern Finland, Kuopio, Finland)

Published: 2026-10-22T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Graph Mining](<https://devfeed.tech/topics/graph-mining.md>), [Graphs](<https://devfeed.tech/topics/graphs.md>), [Encoding](<https://devfeed.tech/topics/encoding.md>)

Tags: [aggregation](<https://devfeed.tech/tags/aggregation.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [atom](<https://devfeed.tech/tags/atom.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [encoding](<https://devfeed.tech/tags/encoding.md>), [features](<https://devfeed.tech/tags/features.md>), [fragment](<https://devfeed.tech/tags/fragment.md>), [functional](<https://devfeed.tech/tags/functional.md>), [gnn](<https://devfeed.tech/tags/gnn.md>), [graph](<https://devfeed.tech/tags/graph.md>), [graph-neural-networks](<https://devfeed.tech/tags/graph-neural-networks.md>), [groups](<https://devfeed.tech/tags/groups.md>), [property](<https://devfeed.tech/tags/property.md>), [range](<https://devfeed.tech/tags/range.md>), [stable](<https://devfeed.tech/tags/stable.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [test](<https://devfeed.tech/tags/test.md>), [times](<https://devfeed.tech/tags/times.md>), [training](<https://devfeed.tech/tags/training.md>)

## AI overview

The paper changes molecular input representation rather than the GNN: BRICS decomposition forms graphs whose nodes are synthetically meaningful fragments, with features summed from their atoms. This lets message-passing GNNs and post-hoc explainers produce functional-group-level attributions. Across four architectures, four tasks, and two explainers with scaffold splits, fragment graphs were roughly five times smaller while matching atom-level test accuracy. Integrated Gradients recovered known pharmacophores above chance for BACE (1.5x chance) and PTR1 (3.2x). In the reported BACE comparison, aggregating atom-level attributions had hit rates that varied above or below the fragment model depending on two arbitrary choices; fragment-level training gave a single convention-free attribution without predictive cost. The evidence provided is an abstract, so further methodological details are unavailable.

## Source excerpt

arXiv:2610.29589v1 Announce Type: new Abstract: Graph neural networks (GNNs) are among the standard tools for molecular property prediction, but the atom-level explanations of their predictions are hard for medicinal chemists to act on. We propose a change of input representation rather than a new model: molecules are encoded as fragment-level graphs from BRICS decomposition, where each node is a synthetically meaningful fragment whose features are the element-wise sum of its atoms' features. Because only the representation changes, any message-passing GNN and any post-hoc attribution method produces explanations at the resolution of functional groups instead of single atoms. Across four architectures, four prediction tasks, and two explainers under scaffold splits, fragment-level encoding matches atom-level test accuracy on roughly five-fold smaller graphs. Integrated-Gradients attributions recover known pharmacophores above chance on two enzyme targets, namely BACE at 1.5 times the chance rate, and PTR1 at 3.2 times. A like-for-like comparison against an aggregated atom-level baseline shows that post-hoc aggregation is not a stable substitute: its BACE hit rate range both exceeds and falls below the fragment model's hit rate depending on two arbitrary aggregation choices, whereas fragment-level training yields a single convention-free attribution at no predictive cost. The code is available at: https://github.com/4D-Lab/frame