IPM-NUS Workshop on Analysis and Application of Protein Interaction Networks
Shahid Beheshti University
17 & 18 November 2008
Full program of the workshop /
A ppt about the School of Computing at NUS.
I gave several invited talks and a tutorial at the workshop above. Here
are the details...
High-throughput experimental methods, such as yeast-two-hybrid and
phage display, have fairly high levels of false positives
(and false negatives). Thus the list of protein-protein interactions
detected by such experiments would need additional wet
laboratory validation. Advances in computational techniques
for assessing the reliability of protein-protein interactions
detected by such high-throughput methods are reviewed in this talk,
with a focus on techniques that rely only on topological information
of the protein interaction network derived from
such high-throughput experiments.
Protein complexes are fundamental for understanding principles
of cellular organizations. However, most protein interactome maps
are still essentially an in vitro scaffold. Further these protein
interactome maps contain a significant amount of noise interactions,
as well as missing many real interactions. It is thus an
important challenge to reliably deduce in vivo protein interactions
and to identify membership in the same protein complexes. In this talk,
we describe recent progress in computational techniques for
protein complex prediction from noisy protein interaction network data.
A central problem of computational biology is the inference of
the function of a protein. The traditional computational approach
to this problem is based on the principle of "guilt by association"
of sequence similarity. This approach works for about 40-60% of
the proteins in a typical proteome. In this talk, we discuss the
inference of function for the other proteins that lack informative
sequence similarity to proteins with known function. We present
guilt by association of common friends --- that two proteins
sharing a large number of common interaction partners are likely to share
a common function. Furthermore, we develop a means to exploit this
property to effectively assign functions to proteins in the absence
of sequence similarity. In order to fully exploit additional
information that is available on some proteins, we also develop an
efficient powerful information fusion technique to infer protein
functions through guilt by association of multiple information types.
One of the most important problems in computational biology is
the reliable assignment of functions to protein sequences. This tutorial
on protein function prediction has two objectives. The first objective
is to show the attendees the breadth of the many different machine
learning and data mining approaches that have been developed for
this problem, and also to highlight the golden thread of ``guilt by
association'' that runs through all of these alternative approaches.
The second objective is to illustrate some of the important
interpretational skills that a good bioinformaticist must have
in order to properly draw conclusions from the output of such computational
algorithms.
Full program of the workshop /
A ppt about the School of Computing at NUS.
I also visited the University of Tehran
Institute of Biochemistry and Biophysics.
Here is the talk I gave there:
Whenever a programmer writes a loop, or a mathematician does a
proof by induction, an invariant is involved. The discovery and
understanding of invariants often underlies problem solving
in many domains. I will discuss here my search for powerful
invariants over the past decade. My search was/is motivated by a
broad spectrum of problems: understanding query languages,
engineering data integration systems, optimising disease treatments,
recognizing DNA feature sites, and discovering reliable patterns.
In the course of my talk, you will discover some of the most
powerful and unexpected invariants in logic, engineering,
medicine, and biology.