Seminars

Realistic images of 3D scenes result from numerical solutions
to the light transport equation. These physically-based
renderings are slower to produce than the default renderings
generated by graphics hardware. As a result, physically-based
illumination is uncommon in 3D medical visualization systems.

When a surgeon is considering treatment options for
a patient with, for example, a brain tumor, the quality
of the data visualization may influence the geometric
inferences made by the surgeon.

This talk described current work in our lab to integrate
physically-based illumination into the clinical workflow
at Harvard Medical School's Surgical Planning Lab,
and reports on our testing of human subjects to quantify
perceptual differences resulting from physically-based
illumination of complex 3D scenes.

Making sense of gene repression in living bacteria requires understanding of the looping properties of DNA in crowded, multi-component systems. The presence of HU, a non-specific binding protein that introduces sharp bends and localized untwisting in double-helical DNA, stabilizes functional repression loops as small as ~65 base pairs. As a first step in the analysis of such looping, we have investigated the effects of HU on the configurational properties of short fragments of ideal B DNA, treating the DNA at the level of base-pair steps and incorporating the known effects of HU on DNA structure. We introduce a new sampling technique to model the non-specific binding of HU on DNA and use Monte-Carlo methods to generate three-dimensional configurations of protein-bound DNA. The presentation will focus on properties of small circular duplexes formed in the presence of HU and the implications of these data for the in-vivo looping properties of DNA.

In a recent article in APS News, John Hopfield has defined physics as "The idea ... that the world is understandable." As a physicist working on biological problems, I pursue this understanding as the ultimate goal. Unfortunately, even the simplest cellular networks, if treated in full detail, may consist of hundreds of thousands of biochemical species and reactions coupling them. They are clearly not understandable, at least in the common sense of this word. The usual approach is to coarse-grain these networks, building compact models that are easier to understand and analyze, yet are functionally equivalent to the original networks. In this talk, I analyze two different approaches to the coarse-graining. A de facto standard is to represent cellular regulatory networks as on/off (Boolean) circuits. I will provide a theoretical argument, supported by some preliminary experimental work from other groups, that such binary picture, while appealing, may be inconsistent with keeping the functional equivalence. As an alternative, I will suggest a more rigorous approach, which proceeds using mathematical tools of quantum and statistical physics, and is capable of coarse-graining, while keeping the molecular signaling function intact. I will illustrate the approach on some simple examples.

Networks which trace the activities and interactions of individuals,
transportation fluxes and population movements on the local and global
scale have been analyzed and found to exhibit large scale
heterogeneity, self-organization and other properties typical of
complex systems. Here we analyze the impact of mobility networks on
the global spreading of emerging infectious diseases. We define a
computational model for the large scale spread of infectious diseases
that integrates the air transportation network with demographic data.
The model is used to study the specific case of the SARS epidemic and
to provide scenario forecasts for pandemic influenza. The effect of
the network complexity on the global spreading pattern of diseases is
then analyzed by exploiting the analogy of metapopulations mechanistic
models with stochastic reaction-diffusion processes.

Powerful imaging devices (ranging from very large telescopes, to
medical radiology, to modern microscopes) usually combine a device
that collects light, or similar radiation measurements, with a
computer that assembles the collected data into images that can be
viewed by scientists and doctors. Although we want the recorded image
to be a faithful representation of the true image scene, every image
is more or less blurry. In image deblurring, the goal is to recover
the original, sharp image by using a mathematical model of the
blurring process. The key issue is that some information on the lost
details is indeed present in the blurred image, but this ``hidden"
information can be recovered only if we know the details of the
blurring process. Because the blurring process is often modeled as a
convolution equation, deblurring is often referred to as
deconvolution. In this talk we describe computational techniques,
including basic filtering approaches as well as state-of-the-art
algorithms, that can be used for deconvolution.

DNA copy number aberration (CNA) is one of the key characteristics of cancer cells. Recent studies demonstrated the feasibility of utilizing high density single nucleotide polymorphism (SNP) genotyping arrays to detect CNA. Compared with the two-color array-based comparative genomic hybridization (array-CGH), the SNP arrays offer much higher probe density and lower signal-to-noise ratio at the single SNP level. To accurately identify small segments of CNA from SNP array data, segmentation methods that are sensitive to CNA while resistant to noise are required. We have developed a highly sensitive algorithm for the edge detection of copy number data which is especially suitable for the SNP array-based copy number data. The method consists of an over-sensitive edge-detection step and a test-based forward-backward edge selection step. Using simulations constructed from real experimental data, the method shows high sensitivity and specificity in detecting small copy number changes in focused regions. The method is implemented in an R package FASeg, which includes data processing and visualization utilities, as well as libraries for processing Affymetrix SNP array data.

We describe a computational and analytical method (EATDT) which makes it feasible to analyze and interpret genome-scale data for disease association mapping. Along the way, we dispel several myths, among them: 1) one cannot use SNPs in whole genome association studies, because the multiple test correction will destroy power, 2) one cannot use haplotypes in association studies since there are so many of them and one cannot determine a priori which ones to test, 3) one cannot use haplotypes in association studies because the multiple test correction will destroy power, 4) common SNPs cannot be used to find rare disease alleles. Our computational method incorporates haplotype data, accounts for multiple testing, and runs very efficiently in real-time. We describe recent extensions of these methods to classical Case-Control data, and suggest methods to improve fine mapping of disease alleles. We test our methods on simulated and real data. We conclude that applying our algorithm to genome-scale SNP data generated by extant technologies can allow for the detection of disease mutations of small affect. Finally, our results show that one can detect both rare and common disease alleles by association, effectively making the debate between the common disease common variant and common disease rare variant hypotheses moot.

In this talk we introduce the topic of image registration and discuss
its applications. We review classical numerical methods to match
different images and discuss the challenges of solving the problem.
In particular, we introduce the idea of adaptive multilevel refinement
as a method for the solution of the problem. We demonstrate that this
approach can lead to a significant computational saving that enable us
to work on future problems.

Most biological systems actively generate forces and movement. These forces are derived from a chemical energy source, and converted to mechanical force by coupling chemistry with molecular structural changes. I will illustrate the interplay of mechanics and chemistry in molecular and cellular contexts with two examples: (i) the processive movement of dimeric molecular motors (e.g. myosin and kinesin) on cytoskeleton tracks and (ii) the contractile motions accompanying dividing bacterial cells. I will discuss coarse-grained modeling strategies for these problems and relate molecular level information with experimentally measurable forces and movements. My aim is to illustrate that continuum mechanics is useful in a molecular setting, but new ideas are necessary to elucidate the mechanisms of biological force generation.

Molecular motors play crucial roles in diverse biological processes ranging from gene replication and muscle contraction to cell division. They convert chemical energy such as ATP hydrolysis into mechanical work. Using myosin motor as an example, we will illustrate how to use computational tools to both understand the mechanism of energy transduction and engineer their mechanical functions. The aim is to demonstrate the capability of computation in elucidating biological significance and guiding novel protein design.