How Can Data Mining Help Bio-Data Analysis?

[Extended Abstract]
Jiawei Han
Department of Computer Science
University of Illinois at Urbana-Champaign
[email protected]

ABSTRACT 2. HOW DATA MINING MAY HELP BIO-

Recent progress in data mining research has led to the de- DATA ANALYSIS?
velopment of numerous efficient and scalable methods for
Here we list a few interesting themes on data mining that
mining interesting patterns in large databases. In the mean
may help bio-data analysis.
time, recent progress in biology, medical science, and DNA
technology has led to the accumulation of tremendous amounts 1. Data cleaning, data preprocessing, and semantic
of bio-medical data that demands for in-depth analysis. The integration of heterogeneous, distributed bio-medical
question becomes how to bridge the two fields, data min- databases.
ing and bioinformatics, for successful mining of bio-medical
Due to the highly distributed, uncontrolled generation and
data. In this abstract, we analyze how data mining may help
use of a wide variety of bio-medical data, data cleaning, data
bio-medical data analysis and outline some research prob-
preprocessing, and the semantic integration of such hetero-
lems that may motivate the further developments of data
geneous and widely distributed biomedical databases, such
mining tools for bio-data analysis.
as genome databases and proteome databases, have become
an important task for systematic and coordinated analysis of
Keywords bio-medical databases. This has promoted the research and
Bio-medical data analysis, data mining, bioinformatics, data development of integrated data warehouses and distributed
mining applications, research challenges federated databases to store and manage the primary and
derived bio-medical data, such as genetic data. Data clean-
1. INTRODUCTION ing and data integration methods developed in data mining,
such as [9; 3], will help the integration of bio-medical data
In the past two decades we have witnessed revolutionary and the construction of data warehouses for bio-medical data
changes in biomedical research and bio-technology and an analysis.
explosive growth of bio-medical data, ranging from those
2. Exploration of existing data mining tools for bio-
collected in pharmaceutical studies and cancer therapy in-
data analysis.
vestigations to those identified in genomics and proteomics
research by discovering sequencing patterns, gene functions, With years of research and developments, there have been
and protein-protein interactions. The rapid progress of bio- many data mining, machine learning, and statistics analysis
technology and bio-data analysis methods has led to the systems and tools available for use in bio-data exploration
emergence and fast growth of a promising new field: bioin- and bio-data analysis. Comprehensive surveys and introduc-
formatics. tion of data mining methods have been compiled into many
textbooks such as [11; 6; 7]. There are also many textbooks
On the other hand, recent progress in data mining research
on bioinformatics, such as [2; 8; 5; 4]. General data mining
has led to the developments of numerous efficient and scal-
and data analysis systems have been constructed for such
able methods for mining interesting patterns and knowledge
analysis, such as SAS Enterprise Miner, SPSS, SPlus, IBM
in large databases, ranging from efficient classification meth-
Intelligent Miner, Microsoft SQLServer 2000, SGI MineSet,
ods to clustering, outlier analysis, frequent, sequential and
and Inxight VizServer. There are also some bio-specific data
structured pattern analysis methods, and visualization and
analysis software systems, such as GeneSpring, Spot Fire,
spatial/temporal data analysis tools.
VectorNTI, COMPASS, and SMA (Statistics for Microar-
The question becomes how to bridge the two fields, data ray Analysis) in R. These tools are evolving as well. For
mining and bioinformatics, for successful data mining in bio- bio-data analysis, it is important to train researchers to mas-
medical data. Especially, we should analyze how data min- ter and explore the power of these well-tested and popularly
ing may help efficient and effective bio-medical data analysis used data mining tools and packages. A lot of routine data
and outline some research problems that may motivate the analysis work can be done using such tools.
further developments of powerful data mining tools for bio-
With sophisticated bio-data analysis tasks, there is much
data analysis. This is the motivation of this talk.
room for research and development of advanced, effective,
and scalable data mining methods in bio-data analysis. Some
interesting topics in this direction are illustrated as follows.

BIOKDD02: Workshop on Data Mining in Bioinformatics (with SIGKDD02 Conference) page 1

3. Similarity search and comparison in bio-data. 8. Privacy preserving mining of bio-medical data.
One of the most important search problems in bio-data anal- Although information exchange is important, hospitals and
ysis is similarity search and comparison among bio-sequences research institutes may still be reluctant to give out precious
and structures. For example, gene sequences isolated from bio-medical data due to confidentiality, liability, and other
diseased and healthy tissues can be compared to identify concerns. Thus it is important to develop privacy preserv-
critical differences between the two classes of genes. This ing data mining methods, such as [1], to maximally protect
can be done by first retrieving the gene sequences from the privacy while achieving effective data mining.
two tissue classes, and then finding and comparing the fre-
quently occurring patterns of each class. Usually, sequences 3. CONCLUSIONS
occurring more frequently in the diseased samples than in
the healthy samples might indicate the genetic factors of the Both data mining and bioinformatics are fast expanding re-
disease; on the other hand, those occurring only more fre- search frontiers. It is important to examine what are the
quently in the healthy samples might indicate mechanisms important research issues in bioinformatics and develop new
that protect the body from the disease. Similar analysis data mining methods for scalable and effective bio-data anal-
can be performed on microarray data and protein data to ysis. We believe that the active interactions and collabora-
identify similar and dissimilar patterns. Moreover, since bio- tions between these two fields have just started and a lot of
data usually contains noise or non-perfect matches, it is im- exciting results will appear in the near future.
portant to develop effective sequential or structural pattern
mining algorithms in the noisy environment, such as that 4. REFERENCES
recently reported in [12].
[1] R. Agrawal and R. Srikant. Privacy-preserving data
4. Association analysis: identification of co-occurring
mining. In SIGMOD’00, pp. 439–450, Dallas, TX, May
bio-sequences or other correlated patterns.
2000.
Currently, many studies have focused on the comparison of
[2] A. Baxevanis and B. F. F. Ouellette. Bioinformatics: A
one gene to another. However, most diseases are not trig-
Practical Guide to the Analysis of Genes and Proteins
gered by a single gene but by a combination of genes acting
(2nd ed.). John Wiley & Sons, 2001.
together. Association and correlation analysis methods can
be used to help determine the kinds of genes or proteins [3] T. Dasu, T. Johnson, S. Muthukrishnan, and
that are likely to co-occur in target samples. Such analysis V. Shkapenyuk. Mining database structure; or how to
would facilitate the discovery of groups of genes or proteins build a data quality browser. In SIGMOD’02, pp. 240–
and the study of interactions and relationships among them. 251, Madison, WI, June 2002.
5. Frequent pattern-based cluster analysis. [4] R. Durbin, S. Eddy, A. Krogh, and G. Mitchison. Bi-
ological Sequence Analysis: Probability Models of Pro-
Most cluster analysis algorithms are based on either Eu-
teins and Nucleric Acids. Cambridge University Press,
clidean distances or density [6]. However, bio-data often
1998.
consists of a lot of features which form a high dimension
space, and it is crucial to study differentials with scaling [5] W. J. Ewens and G. R. Grant. Statistical Methods in
and shifting factors in multi-dimensional space and discover Bioinformatics: An Introduction. Springer-Verlag, New
pair-wise frequent patterns and cluster bio-data based on York, 2001.
such frequent patterns. One interesting study taken mico-
rarray data as examples is in [10]. [6] J. Han and M. Kamber. Data Mining: Concepts and
Techniques. Morgan Kaufmann, 2001.
6. Path analysis: linking genes or proteins to differ-
ent stages of disease development. [7] T. Hastie, R. Tibshirani, and J. Friedman. The Ele-
ments of Statistical Learning: Data Mining, Inference,
While a group of genes/proteins may contribute to a dis- and Prediction. Springer-Verlag, New York, 2001.
ease process, different genes/proteins may become active at
different stages of the disease. If the sequence of genetic [8] A. M. Lesk. Introduction to Bioinformatics. Oxford
activities across the different stages of disease development University Press, 2002.
can be identified, it may be possible to develop pharmaceu- [9] V. Raman and J. M. Hellerstein. Potter’s wheel: An
tical interventions that target the different stages separately, interactive data cleaning system. In VLDB’01, pp. 381–
therefore achieving more effective treatment of the disease. 390, Rome, Italy, Sept. 2001.
Such path analysis is expected to play an important role in
genetic studies. [10] H. Wang, J. Yang, W. Wang, and P. S. Yu. Clustering
by pattern similarity in large data sets. In SIGMOD’02,
7. Data visualization and visual data mining. pp. 418–427, Madison, WI, June 2002.
Complex structures and sequencing patterns of genes and [11] I. H. Witten and E. Frank. Data Mining: Practical Ma-
proteins are most effectively presented in graphs, trees, cubes, chine Learning Tools and Techniques with Java Imple-
and chains by various kinds of visualization tools. Such vi- mentations. Morgan Kaufmann, 2001.
sually appealing structures and patterns facilitate pattern
understanding, knowledge discovery, and interactive data [12] J. Yang, P. S. Yu, W. Wang, and J. Han. Mining long
exploration. Visualization and visual data mining therefore sequential patterns in a noisy environment. In SIG-
play an important role in biomedical data mining. MOD’02, pp. 406–417, Madison, WI, June 2002.

BIOKDD02: Workshop on Data Mining in Bioinformatics (with SIGKDD02 Conference) page 2