-
Interpretable machine learning for high-dimensional trajectories of aging health
Authors:
Spencer Farrell,
Arnold Mitnitski,
Kenneth Rockwood,
Andrew Rutenberg
Abstract:
We have built a computational model for individual aging trajectories of health and survival, which contains physical, functional, and biological variables, and is conditioned on demographic, lifestyle, and medical background information. We combine techniques of modern machine learning with an interpretable interaction network, where health variables are coupled by explicit pair-wise interactions…
▽ More
We have built a computational model for individual aging trajectories of health and survival, which contains physical, functional, and biological variables, and is conditioned on demographic, lifestyle, and medical background information. We combine techniques of modern machine learning with an interpretable interaction network, where health variables are coupled by explicit pair-wise interactions within a stochastic dynamical system. Our dynamic joint interpretable network (DJIN) model is scalable to large longitudinal data sets, is predictive of individual high-dimensional health trajectories and survival from baseline health states, and infers an interpretable network of directed interactions between the health variables. The network identifies plausible physiological connections between health variables as well as clusters of strongly connected health variables. We use English Longitudinal Study of Aging (ELSA) data to train our model and show that it performs better than multiple dedicated linear models for health outcomes and survival. We compare our model with flexible lower-dimensional latent-space models to explore the dimensionality required to accurately model aging health outcomes. Our DJIN model can be used to generate synthetic individuals that age realistically, to impute missing data, and to simulate future aging outcomes given arbitrary initial health states.
△ Less
Submitted 4 January, 2022; v1 submitted 7 May, 2021;
originally announced May 2021.
-
The potential for complex computational models of aging
Authors:
Spencer Farrell,
Garrett Stubbings,
Kenneth Rockwood,
Arnold Mitnitski,
Andrew Rutenberg
Abstract:
The gradual accumulation of damage and dysregulation during the aging of living organisms can be quantified. Even so, the aging process is complex and has multiple interacting physiological scales -- from the molecular to cellular to whole tissues. In the face of this complexity, we can significantly advance our understanding of aging with the use of computational models that simulate realistic in…
▽ More
The gradual accumulation of damage and dysregulation during the aging of living organisms can be quantified. Even so, the aging process is complex and has multiple interacting physiological scales -- from the molecular to cellular to whole tissues. In the face of this complexity, we can significantly advance our understanding of aging with the use of computational models that simulate realistic individual trajectories of health as well as mortality. To do so, they must be systems-level models that incorporate interactions between measurable aspects of age-associated changes. To incorporate individual variability in the aging process, models must be stochastic. To be useful they should also be predictive, and so must be fit or parameterized by data from large populations of aging individuals. In this perspective, we outline where we have been, where we are, and where we hope to go with such computational models of aging. Our focus is on data-driven systems-level models, and on their great potential in aging research.
△ Less
Submitted 26 October, 2020; v1 submitted 2 August, 2020;
originally announced August 2020.
-
Community structure detection and evaluation during the pre- and post-ictal hippocampal depth recordings
Authors:
Keivan Hassani Monfared,
Kris Vasudevan,
Jordan S. Farrell,
G. Campbell Teskey
Abstract:
Detecting and evaluating regions of brain under various circumstances is one of the most interesting topics in computational neuroscience. However, the majority of the studies on detecting communities of a functional connectivity network of the brain is done on networks obtained from coherency attributes, and not from correlation. This lack of studies, in part, is due to the fact that many common…
▽ More
Detecting and evaluating regions of brain under various circumstances is one of the most interesting topics in computational neuroscience. However, the majority of the studies on detecting communities of a functional connectivity network of the brain is done on networks obtained from coherency attributes, and not from correlation. This lack of studies, in part, is due to the fact that many common methods for clustering graphs require the nodes of the network to be `positively' linked together, a property that is guaranteed by a coherency matrix, by definition. However, correlation matrices reveal more information regarding how each pair of nodes are linked together. In this study, for the first time we simultaneously examine four inherently different network clustering methods (spectral, heuristic, and optimization methods) applied to the functional connectivity networks of the CA1 region of the hippocampus of an anaesthetized rat during pre-ictal and post-ictal states. The networks are obtained from correlation matrices, and its results are compared with the ones obtained by applying the same methods to coherency matrices. The correlation matrices show a much finer community structure compared to the coherency matrices. Furthermore, we examine the potential smoothing effect of choosing various window sizes for computing the correlation/coherency matrices.
△ Less
Submitted 31 May, 2018; v1 submitted 14 March, 2018;
originally announced April 2018.
-
Probing the network structure of health deficits in human aging
Authors:
Spencer G. Farrell,
Arnold B. Mitnitski,
Olga Theou,
Kenneth Rockwood,
Andrew D. Rutenberg
Abstract:
We confront a network model of human aging and mortality in which nodes represent health attributes that interact within a scale-free network topology, with observational data that uses both clinical and laboratory (pre-clinical) health deficits as network nodes. We find that individual health attributes exhibit a wide range of mutual information with mortality and that, with a re- construction of…
▽ More
We confront a network model of human aging and mortality in which nodes represent health attributes that interact within a scale-free network topology, with observational data that uses both clinical and laboratory (pre-clinical) health deficits as network nodes. We find that individual health attributes exhibit a wide range of mutual information with mortality and that, with a re- construction of their relative connectivity, higher-ranked nodes are more informative. Surprisingly, we find a broad and overlapping range of mutual information of laboratory measures as compared with clinical measures. We confirm similar behavior between most-connected and least-connected model nodes, controlled by the nearest-neighbor connectivity. Furthermore, in both model and observational data, we find that the least-connected (laboratory) nodes damage earlier than the most-connected (clinical) deficits. A mean-field theory of our network model captures and explains this phenomenon, which results from the connectivity of nodes and of their connected neighbors. We find that other network topologies, including random, small-world, and assortative scale-free net- works, exhibit qualitatively different behavior. Our disassortative scale-free network model behaves consistently with our expanded phenomenology observed in human aging, and so is a useful tool to explore mechanisms of and to develop new predictive measures for human aging and mortality.
△ Less
Submitted 6 June, 2018; v1 submitted 23 February, 2018;
originally announced February 2018.
-
Unifying ageing and frailty through complex dynamical networks
Authors:
Andrew D. Rutenberg,
Arnold B. Mitnitski,
Spencer Farrell,
Kenneth Rockwood
Abstract:
To explore the mechanistic relationships between ageing, frailty and mortality, we developed a computational model in which possible health attributes are represented by the nodes of a complex network. Each node can be either damaged (i.e. a deficit) or undamaged. Damage of connected nodes facilitates further local damage. Our model recovers the known patterns of frailty and mortality without any…
▽ More
To explore the mechanistic relationships between ageing, frailty and mortality, we developed a computational model in which possible health attributes are represented by the nodes of a complex network. Each node can be either damaged (i.e. a deficit) or undamaged. Damage of connected nodes facilitates further local damage. Our model recovers the known patterns of frailty and mortality without any programmed ageing. It helps us to understand how the observed maximum of the frailty index (FI) might arise, and allows us to start to understand how health deficits accumulate. Large model populations allow us to exploit new analytic tools, including information theory. This will allow us to systematically characterize the effects of sudden changes in the health trajectories of individuals and serve as a way to evaluate large clinical and population databases.
△ Less
Submitted 29 June, 2017; v1 submitted 16 June, 2017;
originally announced June 2017.
-
Network model of human aging: frailty limits and information measures
Authors:
Spencer Farrell,
Arnold Mitnitski,
Kenneth Rockwood,
Andrew Rutenberg
Abstract:
Aging is associated with the accumulation of damage throughout a persons life. Individual health can be assessed by the Frailty Index (FI). The FI is calculated simply as the proportion $f$ of accumulated age related deficits relative to the total, leading to a theoretical maximum of $f \leq 1$. Observational studies have generally reported a much more stringent bound, with $f \leq f_{max} <1$. Th…
▽ More
Aging is associated with the accumulation of damage throughout a persons life. Individual health can be assessed by the Frailty Index (FI). The FI is calculated simply as the proportion $f$ of accumulated age related deficits relative to the total, leading to a theoretical maximum of $f \leq 1$. Observational studies have generally reported a much more stringent bound, with $f \leq f_{max} <1$. The value of $f_{max}$ in observational studies appears to be non-universal, but $f_{max} \approx 0.7$ is often reported. A previously developed network model of individual aging was unable to recover $f_{max}<1$ while retaining the other observed phenomenology of increasing $f$ and mortality rates with age. We have developed a computationally accelerated network model that also allows us to tune the scale-free network exponent $α$. The network exponent $α$ significantly affects the growth of mortality rates with age. However, we are only able to recover $f_{max}$ by also introducing a deficit sensitivity parameter $1-q$, which is equivalent to a false-negative rate $q$. Our value of $q=0.3$ is comparable to finite sensitivities of age-related deficits with respect to mortality that are often reported in the literature. In light of non-zero $q$, we use mutual information $I$ to provide a non-parametric measure of the predictive value of the FI with respect to individual mortality. We find that $I$ is only modestly degraded by $q<1$, and this degradation is mitigated when increasing number of deficits are included in the FI. We also find that the information spectrum, i.e. the mutual information of individual deficits vs connectivity, has an approximately power-law dependence that depends on the network exponent $α$. Mutual information $I$ is therefore a useful tool for characterizing the network topology of aging populations.
△ Less
Submitted 5 November, 2016;
originally announced November 2016.
