Computational medicine: translating models to clinical care

Abstract

Because of the inherent complexity of coupled nonlinear biological systems, the development of computational models is necessary for achieving a quantitative understanding of their structure and function in health and disease. Statistical learning is applied to high-dimensional biomolecular data to create models that describe relationships between molecules and networks. Multiscale modeling links networks to cells, organs, and organ systems. Computational approaches are used to characterize anatomic shape and its variations in health and disease. In each case, the purposes of modeling are to capture all that we know about disease and to develop improved therapies tailored to the needs of individuals. We discuss advances in computational medicine, with specific examples in the fields of cancer, diabetes, cardiology, and neurology. Advances in translating these computational methods to the clinic are described, as well as challenges in applying models for improving patient health.

Fig. 1

Function is distributed across multiple biological scales. Physiological systems have feed-forward pathways between successive levels, and feedback pathways that span levels of biological organization. Function is distributed and does not necessarily originate from any one level, such as the gene. As a consequence of this complexity, understanding physiological systems in health and disease can only be achieved through quantitative modeling and cannot be understood using “mental models.”

Fig. 2

Representative modeling approaches and their data requirements. The abscissa lists different approaches to modeling biological systems at different scales (ordinate). “Algebraic” models describe classes of objects in the genome and their relationships [for example, (107)]. “Topological” models describe molecular wiring diagrams [for example, (108)]. “Statistical” models describe molecular networks as the joint probability distribution of molecular concentrations [for example, (35)]. “Dynamical,” mechanistic models describe the spatiotemporal evolution of biological states using ordinary or partial differential equations [for example, (68)]. “Agent-based” models describe physiological system component interactions using rules, and component location and state evolve in time [for example, (109)]. “Geometrical” models describe anatomic shape [for example, (88)]. Representative data types used in these different modeling approaches and at different biological scales are color-coded. EEG, electroencephalography; ECG, electrocardiography; PET, positron emission tomography.

Fig. 3

Statistical learning of molecular networks and disease phenotypes. The panel on the left shows the names of a set L of measured molecular species, where X represents the observed states [for singlen-ucleotide polymorphisms (SNPs) ] or concentrations (for mRNAs and proteins) for an individual. A topology of parts (molecular network) is then constructed, where nodes are the measured species L and edges represent molecular interactions. The rightmost panel shows how a statistical model of disease, expressed as the likelihood of the measurements X, in disease divided by likelihood in normal, is used to make a personalized diagnosis. Building this classifier for diagnosing the individual requires knowing the likely and unlikely states that distinguish health from disease.

Fig. 4

Modeling electrical activity in the infarcted heart. Clinical MR scan of an infarcted patient heart before ablation (treatment) and the corresponding segmentation: healthy (red), gray zone (GZ) of functional but impaired tissue surrounding the scar (green), or scar (yellow). A three-dimensional geometric model of the patient heart was rendered with the epicardium and the infarct border zone semitransparent. An in silico activation map of VT reveals reentry on the left ventricular endocardium. The color code in the bottom right shows electrical activation time.

Fig. 5

Methods of computational anatomy. (A) A sample of hippocampi from a population of healthy subjects (n = 57) and subjects with very mild AD (n = 38) (88). The anatomical template was generated from all hippocampi from this study. Also shown are the distances between nine individual hippocampi selected from the population and the template. (B) Patterns of hippocampal shape change over a 2-year period in healthy elderly subjects and subjects with very mild AD (90). The shape and volume changes revealed using CA methods support the detection of AD onset.