GUEST COMMENTARY | MICHAEL LIEBMAN March 17, 2004 | Systems biology involves the representation and analysis of an intact biological system. Like many of the technological developments over the past 20 years, such as genomics, proteomics, combinatorial chemistry, and bioinformatics, pharma and medical communities hold high hopes that systems biology will help move molecular research closer to the practice of medicine.
BRICKS IN THE WALL: A top-down approach has important implications for understanding human disease.
But given how slowly any of these advances have yielded high impact, it is time to re-investigate how we examine the interface between biology and information technology. I propose that we apply a top-down
approach to systems biology to complement the bottom-up approach that has been followed with these other disciplines.
There are two reasons why we should augment our traditional reductionist approach to science: the complexity of the system to be studied, and the limitation of our existing knowledge.
In the bottom-up approach, we study basic components and integrate the data to detect relevant patterns (e.g., subatomic particles form the nucleus). In the top-down approach, we establish our knowledge of the system, and attempt to disassemble it. A box containing the pieces of a model train set and the fully assembled train exemplify the difference. They are equivalent in composition but not function, because of the hierarchical interactions among the components.
The system of human physiology requires information on all parts and all interactions. In the bottom-up approach, we integrate all known components and interactions to model the system; top-down, we start with the intact system and decompose it into component parts and interactions. The critical difference occurs when all components and interactions are not known.
Many of the current efforts in systems biology look to integrate results of today's scientific technologies responsible for the ubiquitous "data overload." The difficulty resides in converting data into information, and then into knowledge. The initial transition requires data cleansing and data coherency, but turning information into knowledge requires interpreting what the data actually mean, and how they address questions that need to be answered.
The top-down approach in systems biology should focus on refining those questions and optimizing the path to their solution. This requires patient studies in both "normal" and "abnormal" (disease) conditions, to better understand the current practice of medicine and to address the complexity resident in the intact "biological organism." This is particularly relevant as the bottom-up approach generates data that may be much more quantitative than that currently produced through the conventional practice of medicine.
Our understanding of the natural course of most diseases is significantly limited. Rarely does a physician "observe" a patient at the beginning of any disease. A disease should be viewed as a complex set of accumulating changes that result in multiple symptoms or effects. Some symptoms predominate in diagnosis, but they may conceal more significant problems associated with the full complexity of the disease etiology. In addition, two patients with similar symptoms may be at different stages in two different diseases, and patients with differing symptoms may be at different stages of the same disease. We should deal with this problem by collecting temporal data and developing better computational analysis methods to "synchronize" patients along disease vectors.
To identify disease subtypes, synchronization of clinical histories is essential. Such classification of complex disorders can overcome limitations in detecting genetic linkages among groups of patients. Conventionally, diagnostics are not developed using the disease etiology, but rather on the statistical analysis of "normal" populations. Animal models may be limited in studying natural disease progression (e.g., 90 percent of breast cancer is postmenopausal; the mouse model for breast cancer does not undergo menopause). The conversion of data to knowledge requires better methods in text data mining and knowledge representation.
Mendelian inheritance does not guarantee identical disease presentation because of additional genetic variation. Interactions with environment and lifestyle should be treated in terms of dosing (i.e., how much and when patients were exposed to these environmental perturbants). These observations suggest analyzing the patient and the disease before attempting analysis within a genomically based systems biology approach.
We still don't understand what a protein structure represents in defining biological function, let alone physiological function. Proteins are a level of complexity above the genome. While we wait for science to better identify the "first principles in biology," informatics and computer technology can do much to improve patient care. This will only help us define the system from the bottom up, because we will have shown that it can have an impact from the top down.
Michael Liebman is chief scientific officer of Windber Research Institute, in Windber, Pa. E-mail: firstname.lastname@example.org.