......................................................................

NJIT Mathematical Biology Seminar

Tuesday, October 2, 2007, 4:00pm
Cullimore Hall 611
New Jersey Institute of Technology

......................................................................

Combining gene expression and clinical data for diagnosis

Esteban Tabak

Courant Institute of Mathematical Sciences, New York University

Patients with heart transplants have to be monitored regularly after surgery, seeking particularly to detect any sign of rejection of the transplanted heart by the host body. The way this monitoring is performed nowadays involves the analysis of a biopsy of the heart, a very invasive procedure.

On the other hand, present technology permits the gathering of massive amounts of information on the state of the patient in a non-invasive manner, for instance thorugh measuring the differential expression of each gene (for a total of tens of thousands) in a drop of peripheral blood, by means of a microarray. Can we link this genetic expression to the state of the patient's heart, in a confident-enough manner so as to eliminate the need for a biopsy? If the information gathered in the genes proves to be too elusive, can we complement it with more traditional clinical information, such as the age, gender, and medical history of both patient and donor?

The question is, of course, more general: can micro-array technology, combined with clinical data, help us diagnose illness? From the mathematical perspective, this is a question in data-mining: How to extract meaningful information from a massive set of data, particularly when this data has a very broad range of types: gene expression, age, ethnicity, physical fitness...

This work will describe work in progress, carried jointly by applied mathematicians at NYU and Cordoba, Argentina, and cardiologists at Columbia University, seeking efficient algorithms to extract elusive information from massive amounts of data. A novel methodology that is emerging from this line of work involves a nonlinear, iterative extension of a principal component analysis, which enables us to uncover ``hidden variables'' which explain the data in an arbitrary, nonlinear fashion.