Self-supervised multimodal learning for group inferences from MRI data: Discovering disorder-relevant brain regions and multimodal links.

In recent years, deep learning approaches have gained significant attention in predicting brain disorders using neuroimaging data. However, conventional methods often rely on single-modality data and supervised models, which provide only a limited perspective of the intricacies of the highly complex brain. Moreover, the scarcity of accurate diagnostic labels in clinical settings hinders the applicability of the supervised models. To address these limitations, we propose a novel self-supervised framework for extracting multiple representations from multimodal neuroimaging data to enhance group inferences and enable analysis without resorting to labeled data during pre-training. Our approach leverages Deep InfoMax (DIM), a self-supervised methodology renowned for its efficacy in learning representations by estimating mutual information without the need for explicit labels. While DIM has shown promise in predicting brain disorders from single-modality MRI data, its potential for multimodal data remains untapped. This work extends DIM to multimodal neuroimaging data, allowing us to identify disorder-relevant brain regions and explore multimodal links. We present compelling evidence of the efficacy of our multimodal DIM analysis in uncovering disorder-relevant brain regions, including the hippocampus, caudate, insula, - and multimodal links with the thalamus, precuneus, and subthalamus hypothalamus. Our self-supervised representations demonstrate promising capabilities in predicting the presence of brain disorders across a spectrum of Alzheimer's phenotypes. Comparative evaluations against state-of-the-art unsupervised methods based on autoencoders, canonical correlation analysis, and supervised models highlight the superiority of our proposed method in achieving improved classification performance, capturing joint information, and interpretability capabilities. The computational efficiency of the decoder-free strategy enhances its practical utility, as it saves compute resources without compromising performance. This work offers a significant step forward in addressing the challenge of understanding multimodal links in complex brain disorders, with potential applications in neuroimaging research and clinical diagnosis.

Urinary metabolic profiling of asymptomatic acute intermittent porphyria using a rule-mining-based algorithm.

INTRODUCTION: Metabolomic profiling combines Nuclear Magnetic Resonance spectroscopy with supervised statistical analysis that might allow to better understanding the mechanisms of a disease. OBJECTIVES: In this study, the urinary metabolic profiling of individuals with porphyrias was performed to predict different types of disease, and to propose new pathophysiological hypotheses. METHODS: Urine 1H-NMR spectra of 73 patients with asymptomatic acute intermittent porphyria (aAIP) and familial or sporadic porphyria cutanea tarda (f/sPCT) were compared using a supervised rule-mining algorithm. NMR spectrum buckets bins, corresponding to rules, were extracted and a logistic regression was trained. RESULTS: Our rule-mining algorithm generated results were consistent with those obtained using partial least square discriminant analysis (PLS-DA) and the predictive performance of the model was significant. Buckets that were identified by the algorithm corresponded to metabolites involved in glycolysis and energy-conversion pathways, notably acetate, citrate, and pyruvate, which were found in higher concentrations in the urines of aAIP compared with PCT patients. Metabolic profiling did not discriminate sPCT from fPCT patients. CONCLUSION: These results suggest that metabolic reprogramming occurs in aAIP individuals, even in the absence of overt symptoms, and supports the relationship that occur between heme synthesis and mitochondrial energetic metabolism.

Correction to: Urinary metabolic profiling of asymptomatic acute intermittent porphyria using a rule-mining-based algorithm.

The article Urinary metabolic profiling of asymptomatic acute intermittent porphyria using a rule-mining-based algorithm, written by Margaux Luck, Caroline Schmitt, Neila Talbi, Laurent Gouya, Cédric Caradeuc, Hervé Puy, Gildas Bertho and Nicolas Pallet was originally published Online First without open access. After publication in volume [14], issue [1], Citation ID[10] the author decided to opt for Open Choice and to make the article an open access publication. Therefore, the copyright of the article has been changed to © The Author(s) 2018 and the article is forthwith distributed under the terms of the Creative Commons Attribution 4.0 International License ( http://creativecommons.org/licenses/by/4.0/ ), which permits use, duplication, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made. The original article has been corrected.

Rule-Mining for the Early Prediction of Chronic Kidney Disease Based on Metabolomics and Multi-Source Data.

1H Nuclear Magnetic Resonance (NMR)-based metabolic profiling is very promising for the diagnostic of the stages of chronic kidney disease (CKD). Because of the high dimension of NMR spectra datasets and the complex mixture of metabolites in biological samples, the identification of discriminant biomarkers of a disease is challenging. None of the widely used chemometric methods in NMR metabolomics performs a local exhaustive exploration of the data. We developed a descriptive and easily understandable approach that searches for discriminant local phenomena using an original exhaustive rule-mining algorithm in order to predict two groups of patients: 1) patients having low to mild CKD stages with no renal failure and 2) patients having moderate to established CKD stages with renal failure. Our predictive algorithm explores the m-dimensional variable space to capture the local overdensities of the two groups of patients under the form of easily interpretable rules. Afterwards, a L2-penalized logistic regression on the discriminant rules was used to build predictive models of the CKD stages. We explored a complex multi-source dataset that included the clinical, demographic, clinical chemistry, renal pathology and urine metabolomic data of a cohort of 110 patients. Given this multi-source dataset and the complex nature of metabolomic data, we analyzed 1- and 2-dimensional rules in order to integrate the information carried by the interactions between the variables. The results indicated that our local algorithm is a valuable analytical method for the precise characterization of multivariate CKD stage profiles and as efficient as the classical global model using chi2 variable section with an approximately 70% of good classification level. The resulting predictive models predominantly identify urinary metabolites (such as 3-hydroxyisovalerate, carnitine, citrate, dimethylsulfone, creatinine and N-methylnicotinamide) as relevant variables indicating that CKD significantly affects the urinary metabolome. In addition, the simple knowledge of the concentration of urinary metabolites classifies the CKD stage of the patients correctly.

Energetics of endurance exercise in young horses determined by nuclear magnetic resonance metabolomics.

Long-term endurance exercise severely affects metabolism in both human and animal athletes resulting in serious risk of metabolic disorders during or after competition. Young horses (up to 6 years old) can compete in races up to 90 km despite limited scientific knowledge of energetic metabolism responses to long distance exercise in these animals. The hypothesis of this study was that there would be a strong effect of endurance exercise on the metabolomic profiles of young horses and that the energetic metabolism response in young horses would be different from that of more experienced horses. Metabolomic profiling is a powerful method that combines Nuclear Magnetic Resonance (NMR) spectrometry with supervised Orthogonal Projection on Latent Structure (OPLS) statistical analysis. (1)H-NMR spectra were obtained from plasma samples drawn from young horses (before and after competition). The spectra obtained before and after the race from the same horse (92 samples) were compared using OPLS. The statistical parameters showed the robustness of the model (R2Y = 0.947, Q2Y = 0.856 and cros-validated ANOVA p < 0.001). For confirmation of the predictive value of the model, a test set of 104 sample spectra were projected by the model, which provided perfect predictions as the area under the receiving-operator curve was 1. The metabolomic profile determined with the OPLS model showed that glycemia after the race was lower than glycemia before the race, despite the involvement of lipid and protein catabolism. An OPLS model was calculated to compare spectra obtained on plasma taken after the race from 6-year-old horses and from experienced horses (cross-validated ANOVA p < 0.001). The comparison of metabolomic profiles in young horses to those from experienced horses showed that experienced horses maintained their glycemia with higher levels of lactate and a decrease of plasma lipids after the race.