|
|
 |
Keith Duggar
- Current Position:
- Research Staff Member
Functional Genomics and Systems Biology Group
IBM
research, T.J. Watson Research Center
Yorktown Heights, NY 10598
- Contact Information:
- work : (914) 945-2102
cell : (617) 270-4535
mail : duggar@alum.mit.edu
- Education:
- Chemical Engineering, Georgia Institute of Technology, 1996
- Chemical Engineering, Massachusetts Institute of Technology , 2004
- Research Interests:
-
Probability, Physical Modeling, Data Analysis, Bayesian Analysis, Statistical
Analysis, Gene Expression
When new technologies begin to provide data of a kind never seen before,
a great excitement understandably spreads through the eager research community.
However, often this excitement turns to frustration as users of the technology
begin to discover it quirks and warts and realize there is insufficient
physical understanding and a lack of analytical tools that would allow
either correction of these flaws or maximal use of the data as is.
- This is where I try to help. I consider the technology as from a physical
perspective as fully as possible. The I use this physical knowledge to model the
technology and the measures it yields. Finally these model provide the basis for
probabilistic or Bayesian analysis of the data within the context of the
physical model.
- This method has strengths. First, because it is physically based, it is open
to the full consideration and experimental validation that science offers and
demands. The physicality also provides a common foundation upon which
researchers can share and evaluate data. Second, the knowledge built into the
model can help increase the accuracy of inferences and allow quantification of
uncertainty. Finally, a physical model serves as a guide helping us to narrow
the field of possible analyses. This helps to save time and avoid potentially
insidious errors.
- Of course, the method also has weaknesses. First, developing a physical model
is usually very difficult and consumes both time and effort. Second, if the
physical model is very wrong then inferences can actually be made less accurate.
Third, an inaccurate model can lead investigations astray and away from key
evidence that could reveal new understanding.
- This is why Bayesian Analysis and Statistical Analysis are best used in
tandem. Statistical Analysis is essentially and intuitive aid; a free tool for
consolidating information and reducing data dimensionality to help aid the mind
in gaining physical intuition. Indeed, out work relies heavily on both
approaches.
- Work Experience:
-
Harvard Center for Risk Analysis, 2000-2001
- Selected Publications:
-
Physical Modeling and Analysis of Gene Expression Arrays, Ph. D. Thesis
Cohen, J.T., Duggar, K.H., Gray, G.M, Kreindel, S., Gubara, H., Habtemariam,
T., Oryang, D., and Tameru, B. A Simulation Model for Evaluating the Potential
for Spread of Bovine Spongiform Encephalopathy in Animals or to People. Prions
And Mad Cow Disease. Ed. Nummally, B., and Krull, I.S. Marcel Dekker, Inc. New
York, NY.
Cohen, J.T., Duggar, K.H., Gray, G.M, Kreindel, S., Abdelrahman, H.,
Habtemariam, T., Oryang, D., and Tameru, B. (2001) Evaluation of the Potential
for Bovine Spongiform Encephalopathy in the United States. Prepared for the
United States Department of Agriculture.
|
|
