Triantafyllidis, C. | bioRxiv Science

bioRxiv2024

Mathematical Programming and Graph Neural Networks illuminate functional heterogeneity of pathways in disease

We employ a computational framework that integrates mathematical programming and graph neural networks to elucidate functional phenotypic heterogeneity in disease by classifying entire pathways under various conditions of interest. Our approach combines two distinct, yet seamlessly integrated, modeling schemes. First, we leverage Prior Knowledge Networks (PKNs) to reconstruct gene networks from genomic and transcriptomic data. We demonstrate how this can be achieved through mathematical programming optimization and provide examples using comprehensive established databases. We then tailor Graph Neural Networks (GNNs) to classify each network as a single data point at graph-level, using various node embeddings and edge attributes. These networks may vary in their biological or molecular annotations, which serves as a labeling scheme for their supervised classification. We apply the framework to the human DNA damage and repair pathway using the TP53 regulon in a pancancer study across cell-lines and tumor samples to classify Gene Regulatory Networks (GRNs) across different TP53 mutation types. This approach allows us to identify mutations with distinguishable functional profiles which can be related to specific phenotypes, thus providing a data-driven pipeline for genotype-to-phenotype translation. This scalable approach enables the classification of diverse conditions within the multi-factorial nature of diseases and disentangles their polygenic complexity by revealing new functional patterns through a causal representation.

systems biology↗

bioRxiv2022

Reconstructing the functional effect of TP53 somatic mutations on its regulon using causal signalling network modelling

AO_SCPLOWBSTRACTC_SCPLOWThe gene encoding tumor protein p53 (TP53) is the most frequently mutated gene in human cancer. Mutations in both coding and non-coding regions of TP53 can disrupt the regulatory function of the transcription factor, but the functional impact of different somatic mutations on the global TP53 regulon is complex and poorly understood. To address this, we first proceed with a machine learning (ML) approach, and then propose an integrated computational network modelling approach that reconstructs signalling networks using a comprehensive collection of experimental and predicted regulons, and compares their topology. We evaluate both these approaches in a scrutinized pan-cancer analysis of matched genomics and transcriptomics data from 1,457 cell lines (22 cancer types) and 12,531 clinical samples (54 cancer sub-types). Using a ML approach based on penalized generalized linear regression we were able to predict TP53 mutation, but failed to resolve different mutation types. Thus, to infer the impact of different TP53 mutations we compared the topological characteristics of the optimized and reconstructed (upwards of twenty thousand) gene networks and extracted gene signatures for each mutation type using network analysis. We demonstrate that by accounting for TP53 mutation characteristics such as i) mutation type (e.g. missense, nonsense), ii) deleterious consequences of the mutation, or iii) mapping to previously identified hotspots, we can infer a much richer understanding of gene expression regulation, than when simply grouping samples based on their mutation/wild type or gene expression status. Our study highlights a powerful strategy exploiting signalling networks to systematically characterize the functional impact of the full spectrum of somatic mutations. This approach can be applied in general to genetic variation, with clear implications for, but not limited to, the biomedical domain and precision medicine.

systems biology↗