Singh 1
Designing a computational system to predict protein-protein interactions in
Arabidopsis Thaliana
Kamaldeep Singh, School of Engineering and Computer Science, University of the Pacific, Stockton, CA
95211
Yanjun, Qi, School of Computer Science, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213
Mentor: Dr. Judith Klein-Seetharaman, Department of Structural Biology, University of Pittsburgh School
of Medicine, Pittsburgh, PA 15260
Introduction
Proteins are unique amongst organic compounds in supporting every reaction
occurring in biological systems. They can be thought of as molecular machines within
cells performing functions specified by the information encoded in genes. Proteins
participate in diverse biological functions ranging from DNA replication, cell division,
structural support, and signal transduction, among others. Over the last decade enormous
amount of sequencing information has been generated via the Human Genome Project.
However, the genome is merely a parts list which nature uses to assemble proteins. It is
the proteins that are responsible for the tangible features representing life. In order to
recognize and understand the diverse roles of proteins it is first important to understand
the mechanism by which the linear sequences provided by the genome project are
converted into three-dimensional proteins. Determining protein structure is beneficial as
it yields information about its dynamics and ultimately function. Further, it is even more
important to understand the mechanism by which a set of proteins communicate and
collaborate toward a common cellular function. Identifying interactions between proteins
will allow a broader outlook and a more comprehensive understanding of the biological
processes occurring in cells. Also, accurate predictions of protein interaction can be
useful in making implications about the functions of unknown proteins according to their
interacting partners. 1
Singh 2
Protein-protein interaction can be categorized into three subtasks: (1) physical
interaction, (2) interaction amongst proteins in the same complex, and (3) proteins
involved in pathway networks.1 Direct high-throughput experimental methods such as
two-hybrid screens and mass spectrometry can predict the set of interactions amongst
proteins. 1 However, these methods often yield false-positive and false-negative results. 1
Although, it was shown that indirect datasets can also be used in protein interaction
prediction and thus integrating direct high-throughput methods with indirect biological
data sources can lead to more accurate results. 2,3
Predicting protein-protein interactions is one of the most challenging yet motivating
problems of the post-genomic era. Hence, researchers are shifting their focus from the
study of a single protein to that of the entire proteome. Therefore it becomes necessary to
systematically evaluate the different methods that are utilized in predicting protein-
protein interactions. Ultimately the method(s) that perform the best in terms of accurately
predicting protein-protein interaction can be used to model the proteome of a given
organism.
Methods
Recent studies conducted by Yanjun Qi, Ziv Bar-Josepth, and Judith Klein-Seetharaman
employed a combination of three different data sets, two feature encoding styles, and six
different methods.1 Each data set was used to predict a particular protein-protein
interaction subtask. Database of Interacting Proteins (DIP) was used to predict the
physical interaction amongst proteins.1 For predicting proteins involved in the same
complex the Munich Information Center for Protein Sequences (MIPS) complex catalog
Singh 3
was used.1 Lastly, the Kyoto Encyclopedia of Genes and Genomes (KEGG) database was
used to infer protein interaction in pathway networks.1
The way the data is encoded was also varied. Two different types of feature encoding;
Summary and Detailed were considered.1 Summary encoding involved similar
experiments that were combined to produce a single value.1 On the contrary, in the
detailed encoding each experiment was represented separately to provide multiple
values.1
Additionally, six different classifiers, Random Forest (RF), RF similarity-based k-
Nearest-Neighbor (kRF), Nave Bayes (NB), Decision Tree (DT), Logistic regression
(LR), and Support Vector Maching (SVM) were compared in order to highlight the key
differences in their performance.1 These classifiers are all models used to estimate
interactions occurring amongst proteins but they differ from one another in that they each
take a unique approach towards estimating specific types of interactions (physical, co-
complex, co-pathway network). Therefore, different classifiers were considered for each
protein prediction subtask and independent of the prediction subtask it was observed that
the RF classifier performed the best in terms of accurately predicting interactions. 1
Goals
It is the goal of this project to build a computational protein-protein interaction prediction
system for Arabidopsis Thaliana. As mentioned above the Random Forest (RF) classifier
has been shown to be the method of choice and therefore this method will be utilized in
predicting potential protein-protein interactions for this plant. Several genes in
Arabidopsis Thaliana are common with crop and medicinal plants therefore a complete
proteome of Arabidopsis will be of great industrial use.
Singh 4
The project of protein-protein interactions can be further extended to humans. This will
be a major step towards characterizing the interaction of all the proteins. Predicting
protein-protein interactions will enhance the current understanding of diseases and will
pave the way for novel therapeutic approaches.
1
Qi Y, Bar-Joseph Z, Klein-Seetharaman J. Evaluation of different biological data and computational
classification methods for use in protein interaction prediction. Proteins 2006;63:490-500.
2 Ito T, Chiba T, Ozawa R, Yoshida M, Hattori M, Sakaki Y. A comprehensive two-hybrid analysis to
explore the yeast protein interactome. Proc Natl Acad Sci USA, 2001; 10:4569-4574.
3 Utez P, Giot L, Cagney G, Mansfield TA, Judson RS, Knight JR, Lockshon D, Narayan V, Srinivasan M,
Pochart P, Qureshi-Emili A, Li Y, Godwin B, Conover D, Kalbfleisch T, Vijayadamodar G, Yang M,
Johnston M, Fields S, Rothberg JM. A comprehensive analysis of protein-protein interactions in
Saccharomyces cerecisiae. Nature 2000; 403:623-627.
