Robust Physiological Metrics From Sparsely Sampled Networks

Abstract

Physiological and biochemical networks are highly complex, involving thousands of nodes as well as a hierarchical structure. True network structure is also rarely known. This presents major challenges for applying classical network theory to these networks. However, complex systems generally share the property of having a diffuse or distributed signal. Accordingly, we should predict that system state can be robustly estimated with sparse sampling, and with limited knowledge of true network structure. In this review, we summarize recent findings from several methodologies to estimate system state via a limited sample of biomarkers, notably Mahalanobis distance, principal components analysis, and cluster analysis. While statistically simple, these methods allow novel characterizations of system state when applied judiciously. Broadly, system state can often be estimated even from random samples of biomarkers. Furthermore, appropriate methods can detect emergent underlying physiological structure from this sparse data. We propose that approaches such as these are a powerful tool to understand physiology, and could lead to a new understanding and mapping of the functional implications of biological variation.

The New Field of Network Physiology: Building the Human Physiolome

“…It is also important to note current limitations, when one explores uncharted territory through the perspectives of Network Physiology. The progress towards a reliable network-based approach to disease is still limited by the incompleteness of the available data on protein-protein interactions, metabolic networks, information of biological regulatory pathways and organ interactions that are heavily relying on large scale biomedical experiments and streams of continuous physiological signals (Barabási et al, 2011;Cohen et al, 2021). Meanwhile, as research moves towards the dynamic interactome (Przytycka et al, 2010), it would certainly require new advances in temporal and adaptive networks to probe temporal variations in network topology and function.…”

“…Utilizing this new perspective, recent studies have focused on 1) investigating brain-brain network interactions across distinct brain rhythms and locations, and their relation to new aspects of neural plasticity in response to changes in physiologic state; 2) characterizing dynamical features of brain-organ communications as a new signature of neuroautonomic control; 3) establishing basic principles underlying coordinated organ-organ communications, and 4) constructing first dynamic maps of physiological systems and organ interactions across distinct physiologic states (Bashan et al, 2012;Bartsch et al, 2012;Ivanov and Bartsch, 2014;Liu et al, 2015a;Liu et al, 2015b;Bartsch et al, 2015;Lin et al, 2016;Ivanov et al, 2017Ivanov et al, , 2021bDvir et al, 2018;dos Santos Lima et al, 2019;Lin et al, 2020;Rizzo et al, 2020;Ivanov et al, 2021a;Balagué et al, 2020). Pioneering investigations have made first insights into structural and functional connectivity of physiologic networks underlying individual organ systems and their sub-systems (Tass et al, 1998;Bullmore and Sporns, 2009;Gallos et al, 2012;Liu et al, 2015a;Neufang and Akhrif, 2020;Cohen et al, 2021;Cook et al, 2021), and how global behaviors at the organism level, different physiologic states and functions arise out of networked interactions among organ systems to generate health or disease (Bashan et al, 2012;Ivanov and Bartsch, 2014;Karavaev et al, 2020;Pernice et al, 2020;Tecchio et al, 2020;Wood et al, 2020;Zavala et al, 2020;Angelova et al, 2021;Guillet et al, 2021;…”

Sex Differences in the Physiological Network of Healthy Young Subjects

“…Therefore, physiological differences between men and women also occur in systems that are not easily recordable by continuous monitoring, and are instead widely approached through transversal studies of human populations. To take advantage of the wide selection of physiological variables available for transversal studies, correlation matrices in narrow-age cohorts can be used to construct complex inference networks ( Hofer and Sliwinski, 2001 ; Batushansky et al, 2016 ; Barajas-Martínez et al, 2020 , 2021 ; Cohen et al, 2021 ). Complex inference networks allow to find and explore statistical associations from the perspective of network theory, which provides a natural way to describe the relationships between a large set of entities ( Stephens et al, 2018 ).…”

Mahalanobis distance, a novel statistical proxy of homeostasis loss is longitudinally associated with risk of type 2 diabetes

“…The MD of circulating biomarkers has become an alternative means to analyze such variations in high throughput biomarkers and provide a quantifiable proxy of homeostasis loss. Recently, the calculation of MD has been upgraded by replacing the raw biomarker information with PCs [43]. The rationale of replacing the raw biomarkers with PCs is based on the fact that PC analysis could detect underlying processes that might simultaneously regulate the levels of the variables used in the analysis, but may not be directly measurable [44].…”