Introduction: Precision Quantitative Diagnostic Phenotypes will improve clinical trials by (i) helping to target the right drug to the right patient and (ii) elucidating mechanisms of treatment effect. In addition, these diagnostic phenotypes will accelerate precision medicine by synergising Precision Treatment Response Phenotypes as shown in Figure 1. Compute both types of phenotypes by applying the Science of Individuality Measurement Algorithm (SIMA) to multivariate time series data. Such data are becoming ubiquitous with sensors, various monitoring devices, and functional imaging.
Figure 1. The Closed Loop Patient-Centered Precision Medicine Learning System Enabled by SIMA
See a companion Harvard Forum submission that includes more detail about SIMA but focuses on Precision Treatment Response Phenotypes. This submission focuses on diagnostic phenotypes.
Precision Quantitative Diagnostic Phenotypes provide an alternative to the approach taken by classical diagnostic taxonomies such as the Diagnostic and Statistical Manual of Mental Disorders (DSM) and the International Classification of Diseases.
Precision Quantitative Diagnostic Phenotypes compared to classical taxonomies:
- Are quantitative as distinct from being more qualitative and categorical. Cluster quantitative phenotypes.
- Measure order and disorder per se as distinct from signs and symptoms of disorder.
- Capture dynamics compared to being more static and fixed.
- Are derived from multivariate time series as distinct from impressions and data collected at one or just a few points in time such as during clinic visits.
- Are computed from data and more scientific.
Precision Quantitative Diagnostic Phenotypes apply best to complex chronic functional disorders and diseases such as neuropsychiatric, cardiovascular, endocrine, and immune system disorders. Such disorders tend to emerge, vary and fluctuate in level, adapt to treatments and other exposures, and progress or sometimes regress over time. Multivariate time series can provide orders of magnitude more information for diagnosing and understanding individual Complex Adaptive Systems scientifically compared to data collected at one or just a few times.
Figure 2 is a cartoon representation of Precision Quantitative Diagnostic Phenotypes that quantify functional connectivity. Figure 2 represents one person with a directed network graph with action nodes and edges. Action nodes can vary and fluctuate in level over time. SIMA applies to multivariate time series about action nodes.
Figure 2. Precision Diagnostic Phenotypes Quantify Functional Network Connectivity
Figure 2 represents great complexity with only a few action nodes. A real person has billions of nodes, including individual neurons and types of biologically active molecules, that interact over time. The key to Figure 2 identifies various types of interactants. "Action omics" refers to examples such as gene expression, protein, lipid, carbohydrate, and metabolite levels as distinct from germline genetic characteristics that identify the individual person from conception to long after death. All nodes are internal to the individual to show function for diagnosis as distinct from response to the environment including treatments and agency of the person on that person's environment. These three - function, response, and agency - illustrate the tripartite definition of "work" shown in Figure 3 of the companion submission.
Functional and effective connectivity are major topics of investigation for brains. Figure 2 extends this approach to the entire body. The use of SIMA to compute Precision Quantitative Diagnostic Phenotypes contrasts with classical tools for investigating connectivity - tools such as correlation coefficients.
Some nodes and clusters of nodes in Figure 2 represent spatially localized body parts such as brain, spinal cord, heart, stomach, gut, sex organs, and muscles as well as sense organs in the head and on the skin. Node sizes represent average levels of activity for nodes over time for the particular person.
The moderately sized node near the elbow on the right side of the person in Figure 2 represents substances in bodily fluids such as blood often drawn from an arm. Such laboratory variables are far less spatially localized than, for example, activity levels in brain regions.
In addition, consider the person in Figure 2 to be a node that represents a person as a whole. Some items in the key represent the person as a whole. The whole person is more than the sum of its parts.
Coordinated action is an emergent system property that can not be isolated to any particular body part or node. Dyspraxia is a disorder of coordination. Movement of the whole body can be more or less coordinated in multiple ways. Let the wide red outward-bound arrows in the head of Figure 2 represent brain control of coordinated motion or other activity of the person as a whole. In addition, brain and hormonal activity can be more or less coordinated.
Big Idea – Measure coordinated action as an emergent system property with sets of scores computed by SIMA.
Figure 2 includes arrows to represent Interaction-over-Time (IoT scores) computed by SIMA. Each arrow quantifies the amount and direction of evidence (positive or negative) for an interaction over time. Each IoT score is mathematically standardized using the person's own time series data to have mean = 0 and standard deviation = 1 unless 0 is the only possible score. IoT scores with values of 0 indicate no evidence for an interaction-over-time. Expect IoT scores with values of about 0 in the long run by chance alone. Positive IoT Scores are coded green in Figure 2 and quantify the amount of evidence that higher levels of one node are associated over time with higher levels of a second node. Red arrows quantify the opposite. Arrow width represents the magnitude of an IoT score.
Figure 2 shows only pairwise interactions over time. SIMA is capable of quantifying more complex interactions over time.
Every person would be different to some degree when represented by Figure 2.
Let Figure 2 represent a healthy person. Figure 3 represents a healthy person together with what happens to a person with a major health problem (being comatose) and being dead. Consider disorders and diseases to be somewhere on a spectrum between being healthy and being dead.
Figure 3. Healthy, Comatose, and Dead
Precision Quantitative Diagnostic Phenotypes computed by SIMA from multivariate time series appear to measure that which distinguishes being alive and healthy from being dead.
Consider this thought experiment (Gedankenexperiment) to help appreciate the importance of this distinction. Imagine measuring levels of thousands of substances and biomarkers from samples of blood collected moments before and moments after death by guillotine. What is the difference? Does this difference capture the difference between being alive and being dead?
Precision Quantitative Diagnostic Phenotypes appear to be a new approach for creating disease taxonomies and biomarkers for chronic functional disorders. Such taxonomies might help capture the difference between being alive and healthy, being disordered and diseased, and being dead.
The comatose person in Figure 3 shows substantially degraded action in the nodes (many smaller nodes) as well as fewer and weaker interactions over time (fewer and smaller arrows, especially in the brain).
The dead person is represented with no active nodes and no arrows. Many parts would still be there - a dead brain, a dead heart, etc. However, action, interaction, and coordinated action have ceased.
Big Idea – SIMA helps extend precision medicine beyond molecular medicine.
Might it behove us to do more to actually measure that which helps distinguish the living from the dead and the types and degrees of disorder that come between?
Simple Hormone Demonstration: Figure 4 shows time series (action nodes) for two different reproductive hormones with 143 repeated measurements taken every five minutes for about 12 hours for one individual. These data happen to be from one ewe. New technologies and devices are making it easier to collect time series. These data represent a biological mechanism.
Figure 4. Time Series for Two Reproductive Hormones
SIMA quantifies evidence for interaction-over-time or coordination of action. Figure 5 shows summary results for the data in Figure 4.
Figure 5. Summary Interaction-over-Time Score Results as Functions of Four SIMA Analysis Parameters
The most extreme IoT summary score in Figure 5 has a magnitude of 76.028 standard deviations - Bagnes when computed with SIMA. This score helps quantify and standardize what the mind's eye can see in Figure 4.
This interaction-over-time does not appear to be linear. Part B of Figure 5 suggests that higher levels of GnRH increase levels of LH only up to the level indicated by the most extreme score. Methods that assume linearity are of limited value for investigating nonlinear phenomena.
Part C of Figure 5 for "Time Delay" indicates that the interactions-over-time are rapid, especially for GnRH. Assessment of delay is limited by the temporal resolution of the data (5 minutes for Figure 4). Part D indicates substantial persistence of apparent effect at non-zero levels of persistence, especially for GnRH to LH.
Use SIMA to help assess the temporal criterion of causal relationships as with non-experimental data shown in Figure 4. Notice that in Part C of Figure 5 that there is about a 55 standard deviation difference at a delay of 5 minutes for "GnRH to LH" compared to "LH to GnRH." This, together with other results in Parts C and D of Figure 5, suggest that GnRH drives LH but not the opposite. This result is in accord with biological knowledge and helps validate SIMA.
Figure 5 helps quantify a biological mechanism in considerable detail starting at the level of one individual. This mechanism can vary in different ways and to different degrees. This is part of a mechanistic Precision Quantitative Diagnostic Phenotype as indicated in Figure 1.
The first sentence of the "Introduction" says that Precision Quantitative Diagnostic Phenotypes will help elucidate mechanisms of treatment effect as in clinical trials. Here's how. Suppose that one has need to investigate a drug as a potential GnRH agonist or antagonist. Do so by investigating the effects of the drug on results such as those shown in Figure 5.
Simple Brain Demonstration: Figure 2 is an extension of borrowed Figure 6 both in scope (from brain to full person) and in terms of the capabilities of tools used to investigate functional connectivity.
Figure 6. Functional Connectivity as Illustrated for the Human Brain Project
SIMA is being introduced as a new tool to quantify and elucidate functional connectivity as illustrated by both Figures 2 and 6. SIMA and other tools such as correlation coefficients do need to be investigated and evaluated in terms of capabilities such as those described for Figure 5. SIMA has important additional capabilities that remain to be presented.
Functional brain imaging can provide multivariate time series for thousands or hundreds of thousands of small brain regions or voxels (action nodes). Blood oxygen level dependent functional Magnetic Resonance Imaging (fMRI) often provides data with a temporal resolution of about 2 seconds. Figure 7 shows such data for seven brain regions.
Figure 7. Functional Brain Imaging Data for Seven Action Nodes.
Figure 8 shows the highest level summary of IoT scores that result from applying SIMA to the data in Figure 7. Brain experts are invited to consider these results. Use SIMA to drill down into these summary results as described for Figure 5 about the two hormones.
Figure 8. Summary Functional Connectivity Scores for Seven Brain Regions
Consider the table in Figure 8 to be a small portion of an Action Coordination Profile for one person's brain. This illustrates how to compute Precision Quantitative Diagnostic Phenotypes as an alternative classical taxonomies such as DSM. Investigate mechanisms of treatment effect by quantifying how drugs affect such profiles.
In summary, SIMA deserves to be investigated as a powerful new tool to help advance precision medicine.