TL;DR
This study presents a scalable graph neural network-based method that uses structural interactomics and protein interaction networks to predict the inheritance mode and molecular mechanisms of genetic diseases across the proteome.
Contribution
It introduces a novel graph-of-graphs approach combining structural data and network topology for disease inheritance prediction.
Findings
Accurately predicts inheritance modes for autosomal disease variants.
Classifies dominant-associated proteins by functional effect.
Provides a proteome-wide scalable prediction framework.
Abstract
Genetic diseases can be classified according to their modes of inheritance and their underlying molecular mechanisms. Autosomal dominant disorders often result from DNA variants that cause loss-of-function, gain-of-function, or dominant-negative effects, while autosomal recessive diseases are primarily linked to loss-of-function variants. In this study, we introduce a graph-of-graphs approach that leverages protein-protein interaction networks and high-resolution protein structures to predict the mode of inheritance of diseases caused by variants in autosomal genes, and to classify dominant-associated proteins based on their functional effect. Our approach integrates graph neural networks, structural interactomics and topological network features to provide proteome-wide predictions, thus offering a scalable method for understanding genetic disease mechanisms.
Peer Reviews
Decision·ICLR 2025 Conference Withdrawn Submission
The main strengths of the paper are the following: 1) The paper is well-written and easy to follow. 2) The problems addressed in the paper are relevant 3) It is extremely interesting to have a methodology able to address both MOI and functional effects prediction instead of needing to rely on two different strategies for the two tasks. 4) The bioinformatics-related work and processing is accurate. 5) The figures provided help convey the message of the authors more effectively.
The main weaknesses of the paper are the following: 1) The authors did not provide any code. This impinges on the reproducibility and further evaluation of their methods and results. 2) From the methodological point of view, there seems to be not much novelty. The authors use established architectures "out-of-the-box" to tackle the proposed tasks. 3) It seems that the authors did not perform any parameter tuning on their models. Additionally, no information on the hyperparameters used in the
The integration of information at multiple scales is of interest.
The authors do not present a method capable to integrate information at various scales but rather work independently at each scale without exploiting any form of communication between scales.
The graph of graphs idea to predict the mode of inheritance of diseases is novel. The methods are described in great detail and with persuasive experiments.
My biggest concern is that although this work seems to be relevant for predicting mode of inheritance and classifying functional effects, its contribution to deep learning models in the application domain of biology is insufficient. It is well known that GCN, GIN,GAT are three very classical GNN models. And, the graph in graphs idea is also similar to the idea of the paper [1]. So, as far as ICLR is concerned, I think this may not be a notable paper for the community. Also, I would suggest th
Code & Models
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
