Clinical diagnosis of metabolic disorders using untargeted metabolomic profiling and disease-specific networks learned from profiling data.

Untargeted metabolomics is a global molecular profiling technology that can be used to screen for inborn errors of metabolism (IEMs). Metabolite perturbations are evaluated based on current knowledge of specific metabolic pathway deficiencies, a manual diagnostic process that is qualitative, has limited scalability, and is not equipped to learn from accumulating clinical data. Our purpose was to improve upon manual diagnosis of IEMs in the clinic by developing novel computational methods for analyzing untargeted metabolomics data. We employed CTD, an automated computational diagnostic method that "connects the dots" between metabolite perturbations observed in individual metabolomics profiling data and modules identified in disease-specific metabolite co-perturbation networks learned from prior profiling data. We also extended CTD to calculate distances between any two individuals (CTDncd) and between an individual and a disease state (CTDdm), to provide additional network-quantified predictors for use in diagnosis. We show that across 539 plasma samples, CTD-based network-quantified measures can reproduce accurate diagnosis of 16 different IEMs, including adenylosuccinase deficiency, argininemia, argininosuccinic aciduria, aromatic L-amino acid decarboxylase deficiency, cerebral creatine deficiency syndrome type 2, citrullinemia, cobalamin biosynthesis defect, GABA-transaminase deficiency, glutaric acidemia type 1, maple syrup urine disease, methylmalonic aciduria, ornithine transcarbamylase deficiency, phenylketonuria, propionic acidemia, rhizomelic chondrodysplasia punctata, and the Zellweger spectrum disorders. Our approach can be used to supplement information from biochemical pathways and has the potential to significantly enhance the interpretation of variants of uncertain significance uncovered by exome sequencing. CTD, CTDdm, and CTDncd can serve as an essential toolset for biological interpretation of untargeted metabolomics data that overcomes limitations associated with manual diagnosis to assist diagnosticians in clinical decision-making. By automating and quantifying the interpretation of perturbation patterns, CTD can improve the speed and confidence by which clinical laboratory directors make diagnostic and treatment decisions, while automatically improving performance with new case data.

Correction: CTD: An information-theoretic algorithm to interpret sets of metabolomic and transcriptomic perturbations in the context of graphical models.

[This corrects the article DOI: 10.1371/journal.pcbi.1008550.].

CTD: An information-theoretic algorithm to interpret sets of metabolomic and transcriptomic perturbations in the context of graphical models.

We consider the following general family of algorithmic problems that arises in transcriptomics, metabolomics and other fields: given a weighted graph G and a subset of its nodes S, find subsets of S that show significant connectedness within G. A specific solution to this problem may be defined by devising a scoring function, the Maximum Clique problem being a classic example, where S includes all nodes in G and where the score is defined by the size of the largest subset of S fully connected within G. Major practical obstacles for the plethora of algorithms addressing this type of problem include computational efficiency and, particularly for more complex scores which take edge weights into account, the computational cost of permutation testing, a statistical procedure required to obtain a bound on the p-value for a connectedness score. To address these problems, we developed CTD, "Connect the Dots", a fast algorithm based on data compression that detects highly connected subsets within S. CTD provides information-theoretic upper bounds on p-values when S contains a small fraction of nodes in G without requiring computationally costly permutation testing. We apply the CTD algorithm to interpret multi-metabolite perturbations due to inborn errors of metabolism and multi-transcript perturbations associated with breast cancer in the context of disease-specific Gaussian Markov Random Field networks learned directly from respective molecular profiling data.

Histoepigenetic analysis of the mesothelin network within pancreatic ductal adenocarcinoma cells reveals regulation of retinoic acid receptor gamma and AKT by mesothelin.

To enable computational analysis of regulatory networks within the cancer cell in its natural tumor microenvironment, we develop a two-stage histoepigenetic analysis method. The first stage involves iterative computational deconvolution to estimate sample-specific cancer-cell intrinsic expression of a gene of interest. The second stage places the gene within a network module. We validate the method in simulation experiments, show improved performance relative to differential expression analysis from bulk samples, and apply it to illuminate the role of the mesothelin (MSLN) network in pancreatic ductal adenocarcinoma (PDAC). The network analysis and subsequent experimental validation in a panel of PDAC cell lines suggests AKT activation by MSLN through two known activators, retinoic acid receptor gamma (RARG) and tyrosine kinase non receptor 2 (TNK2). Taken together, these results demonstrate the potential of histoepigenetic analysis to reveal cancer-cell specific molecular interactions directly from patient tumor profiles.