Graph neural networks
Does background biological knowledge help predict deadly gene pairs?
Open access · cc by · source: Europe PMC
Feeding a graph neural network a knowledge graph of pathways, diseases and drugs improved prediction of synthetic-lethal gene pairs compared with methods that only saw known gene pairs.
Study at a glance
- Design
- Computational / modelling — Link prediction on SynLethDB gene pairs; KG4SL (knowledge-graph message passing on gene-specific subgraphs) versus 10 baselines using AUC, AUPR and F1, plus sensitivity, convergence and knowledge-graph ablation analyses.
- N
- N=72804 · Balanced dataset of gene pairs: known synthetic-lethal pairs from SynLethDB plus an equal number of randomly drawn unknown pairs.
- Population
- Human gene pairs from SynLethDB and the SynLethKG knowledge graph
- Outcome
- AUC, AUPR and F1 for predicting synthetic-lethal gene pairs
Structured fields used in claim comparison tables when every cited study has a complete layer.
Key findings
KG4SL beat every baseline; against the runner-up, a matrix-factorisation method using gene-ontology similarity, it improved AUC, AUPR and F1 by a few percentage points. It reached an AUC of 0.9470 and AUPR of 0.9564. A knowledge-graph-only embedding performed worst, a GCN on the gene-pair graph alone did much better, and combining the two sources did better still, with the end-to-end KG4SL on top.
Methodology
Synthetic lethality means two genes whose joint loss kills a cell, which is useful for cancer drug targets. The authors built KG4SL, which for each gene samples a subgraph from a knowledge graph linking genes to pathways, biological processes, diseases, compounds and more, passes messages along it, and scores gene pairs. They compared it with 10 baselines spanning matrix factorisation, random-walk embeddings and standard GNNs, and ran an analysis comparing models using only the knowledge graph, only the gene-pair graph, or both.
Limitations
Negative examples were random unknown pairs rather than confirmed non-lethal pairs, so some negatives may be true synthetic-lethal pairs and random negatives can make the task easier than real discovery. The baselines were evaluated with 5-fold cross-validation whereas KG4SL used a single 8:1:1 split, so the comparison is not perfectly like-for-like, and no significance tests are reported. No predicted gene pairs were tested in the lab, and the model ignores cell-line or tissue context, which the authors note matters for real synthetic lethality.
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?
- Does letting each node choose its own depth fix GNN over-smoothing?