DIMACS
DIMACS REU 2013

General Information

2013-06-17-120545
Student: Bryan Karlovitz
Office: CoRE 444
School: West Chester University
E-mail: bk731036 [dot] wcupa [at] edu
Project: Optimization of Personalized Cancer Therapy

Project Description

Years of experience in cancer therapy has taught us that using combinations of drugs for personalized treatment leads to better response rates. However, designing personalized therapies is challenging because of the many drugs available for treatment and the many tumor properties that can inform treatment. We formulate personalized combinatorial therapy as a constrained optimization problem, where markers (tumor properties) and treatment decision rules are assigned to drugs to achieve a high treatment response rate while using drug combinations of minimal size. The goal of this project is to design heuristic algorithms to solve this optimization problem.


Weekly Log

Week 1:
I spent the first week getting familiar with the project. We would like to be able to look at data from clinical trials and, for a given drug, predict the biomarkers that should be used to inform treatment with that drug. On Friday I gave the first presentation.
Week 2:
I spent most of the second week thinking about the problem definition. I tried some simple heuristic methods on simulated data sets.
Week 3:
I had some success this week using a technique to remove a small fraction of biomarkers from a data set. I do a regression to predict the response to a drug given biomarker status in the cancer and then rank the biomarkers according to p value. Then I throw away the worst (something like the bottom 5-10%) and repeat as needed.
Week 4:
I spent most of this week looking for relevant papers. I had a very helpful conversation with a professor at WCU. It turns out there is some literature on predicting extreme non-responders to depression treatment and this problem is similar to mine. Next week I'll try some of the approaches I read about.
Week 5:
Tested more complicated data sets, worked on some R code to look for marker clustering in response and no response groups. I also continued my literature search.
Week 6:
I continued work on the clustering and began writing my report.
Week 7:
Final presentations were on Friday, so I spent the week preparing my slides and doing some writing for my report.
Week 8:
Worked on my final report.

Presentations


Additional Information