Graph neural networks
Does letting each node choose its own depth fix GNN over-smoothing?
Open access · cc by · source: Europe PMC
Letting each drug or protein node decide how far to gather information from its neighbours kept graph features distinct and slightly improved drug-target prediction.
Study at a glance
- Design
- Computational / modelling — Graph of drugs and protein targets with learned node features; node-dependent smoothing depth, then a gradient-boosted tree classifier; 10-fold cross-validation against five published methods.
- N
- N=1923 · 1923 known drug-target interactions among 549 drugs and 424 targets, plus an equal number of randomly paired negatives (a 1:10 imbalanced setting was also run).
- Population
- Drugs and protein targets from DrugBank 3.0 and HPRD
- Outcome
- AUC, AUPR and F1 for predicting drug-target interactions; mean absolute distance between node embeddings as an over-smoothing measure
Structured fields used in claim comparison tables when every cited study has a complete layer.
Key findings
iGRLDTI had the best AUC and AUPR, but margins over the strongest competitors were small (about one to four percentage points on balanced data). In a harder 1:10 imbalanced setting its AUPR advantage was larger, up to 25.5% over one baseline. Ablations showed that using the graph at all helped a lot over raw features, and the adaptive depth added a further gain of a few percent over a standard GNN; its embeddings were also measurably less similar to each other, meaning less over-smoothing.
Methodology
The authors built a network linking 549 drugs and 424 protein targets through 1923 known interactions, giving each drug a feature vector from its chemical structure and each protein one from its sequence. Instead of using the same number of message-passing steps for every node, their method iGRLDTI picks a node-specific depth (node-dependent local smoothing) and averages features up to that depth; a gradient-boosted decision tree then classifies drug-target pairs. They compared it with five published methods under 10-fold cross-validation and ran ablations.
Limitations
The benchmark is small and old, and negative examples were random unlabelled pairs, some of which may be true but undiscovered interactions. Most gains over the best baselines are only around one to two points and come from a single dataset, so generality is untested. The case-study validations use later DrugBank versions, not new lab experiments, and the authors note that nodes lacking biological features, such as newly discovered drugs, will be predicted poorly.
How this study connects
Role on claims
Each row is a claim on a concept or method page where this paper supports, challenges, or qualifies the statement. Roles are hand-checked — not a model guess.
Network context adds information beyond node features.
In drug-target and synthetic-lethality prediction, using graph structure helped over raw features (iGRLDTI), and combining a knowledge graph with the gene-pair graph beat either alone (KG4SL, AUC 0.947).
Evidence for the claim as stated.
Gains over strong baselines were often small (about one to four points for iGRLDTI; small on KIBA for iNGNN-DTI) and came from benchmarks with randomly sampled negatives, single datasets or unequal evaluation protocols, while the protein-structure GNNs showed larger margins.
Evidence for the claim as stated.
Open questions
Tensions this paper is part of
From concept pages' “where studies disagree.” Disagreement means the same question; scope means different assays, populations, or outcomes.
Gains over strong baselines were often small (about one to four points for iGRLDTI; small on KIBA for iNGNN-DTI) and came from benchmarks with randomly sampled negatives, single datasets or unequal evaluation protocols, while the protein-structure GNNs showed larger margins.
Related papers in this topic
Same topic cluster — not a recommendation engine.
- Can spreading signals through a protein network find disease genes?
- Can a known gene network make microarray classifiers interpretable?
- Can a graph neural network sort unknown phage DNA into families?
- Can self-supervised learning predict how mutations change binding?
- Can a graph neural network judge predicted protein shapes?