Co-PI: Tak Igusa, Civil Engineering, WSE
Using techniques grounded in data mining and machine learning, we are developing frameworks for predicting the onset of chronic disorders using data from clinical and bimolecular screening tests. Although applied here to autoimmune disorders, the approach is readily extendable to the predictive screening of other types of disorders.
The objective of this project is to develop robust modeling frameworks for predicting/characterizing the onset of chronic orders using data obtained from clinical/bimolecular screening tests. Drawing upon an unprecedented dataset obtained by researchers in the Johns Hopkins Medical Institute Department of Rheumatology, we will develop and train algorithms that identify the parameters most closely related to current (or future) onset of autoimmune disorders. Although applied here to scleroderma (an autoimmune disorder affecting the skin), the approach is general and will be applicable to a host of other disorders.
For the past three years in the Division of Rheumatology, a data set for approximately 200 scleroderma patients has been collected that includes approximately 120 clinical variables and 150 refined cellular phenotyping variables, obtained through clinical and laboratory tests at six-month intervals. Such a high-dimensional, longitudinal data set on a single auto-immune disease has never been collected under a single clinical setting, and affords a tremendous opportunity for medical science.
The first research task is to use machine learning and other data classification approaches to extract information from this data set that can be used to better understand the characteristics of patients that respond to each type of treatment. This work would require close collaboration with the clinicians, not only to learn the vocabulary, but also to gain insights into the relationships between the data, patients and their diseases.
In chronic diseases, the tissue that is under stress or attack is typically considered to be the passive responder. The hypothesis here is that there is an interaction that resembles a resonance phenomenon. The tissue under attack responds in a manner that causes an increase in activity or effectiveness of the attacking agent, forming a feedback loop that aggravates the attack. Resonance implies some type of tuning wherein a characteristic of the system is matched to the excitation force, resulting in amplified response. Changes in the system that can displace or alter, even slightly, this characteristic would result in a detuned state, thereby breaking the resonance effect and substantially reducing the response. In the chronic disease problem, such as scleroderma, the goal is to determine the biochemical mechanisms that determine resonance so that appropriate interventions can be developed to detune the system and break the resonance effects. It is believed that the tissue itself would contain the information about the key structure of the resonance mechanics, which can be thought of as resonance nodes.