Conclusions
Abstract:
Conducting contemporary biomedical science is an art. Akin to food preparation, the bioinformatician needs to minimize the alteration of data (the food ingredients), but maximize its potential new knowledge and the visual presentations to biology (taste, flavor and aroma). Conceptual frameworks and adaptive computational models are two resources for elevating biomedical science to new heights.
7.1 Change and move forward
In current times, we are surrounded by fast advancing technology, such as smartphones and electronic commerce, which is drastically changing our lifestyle. In contrast, the general public perceives the improvements in medicine as relatively slow. Many unmet medical needs still remain, particularly in the treatment of serious diseases. Despite daily progress in basic biomedical science, lack of ground-breaking clinical applications prevents the lay person from perceiving any progress. The chasm between public expectations and scientific progress has developed into:
1. scientists being aware of their social responsibilities and devoting more effort outside their lab to present their findings to the general public;
2. the employment of a more global, holistic view and utilization of more molecular details for addressing higher level medical questions pertinent to our daily life;
3. the translation of biomedical knowledge from bench to bedside for creating clinically useful applications.
7.2 Presentation, presentation, presentation
The analysis of data of complex biomedical systems is akin to cooking, where the data are raw ingredients. The purpose of cooking is to release and balance the original taste and flavor of the ingredients, by food preparation and seasoning. It is the ingredients (i.e. the data) that carry the important biological clues. The responsibility of the bioinformatician is to present the clues truthfully, while maximizing presentation. We have discussed many useful presentations such as the Manhattan plot, the volcano plot, the Kaplan–Meier plot, the principal component plot and multi-scale genome browsers. These skills can be used flexibly to present the data to convey the message to the audience.
7.3 Domain knowledge plus adaptivity
With the availability of high-throughput technologies and fast accumulating data, it is imperative to increase the throughput of knowledge generation and clinical applications, based on a high concentration of insights. In different parts of this book, we show that adaptive models are critical for the exploration of a deluge of data scattered in a wide and non-linear search space. Examples of adaptive models include the genetic algorithm used for genomic variant analysis; k-means clustering for forming gene modules from RNA level data; principal component analysis for the analysis of sample clustering; and swarm intelligence for systems biomedical science. By letting the data speak for themselves, we can circumvent the limitations resulting from our linear thinking, distil the essence of matter, and achieve meaningful results in a shorter time.
The fairytale of Snow White and the Seven Dwarfs, told by the Grimm Brothers, is a well-known bedtime story. The queen often asked the magical mirror: “Mirror, mirror on the wall, who in the land is the fairest of all?” When we are confronted with the messy, tremendous volumes of biomedical data, we share the same feeling and are urged to ask the computer agents: “Computer, computer, when can the hidden valuable knowledge emerge from the data?” We hope that the computer is wise enough to give us a crisp and insightful answer after intensive computation of the messy data. But most of the time, a computer agent is more like a hard working but relatively inexperienced student who needs to be coached. A computer agent is willing to work even in the middle of the night, yet has difficulty in telling apart trivial observations from real insights. At the end of the day, we still need to provide guidance and directions to computer agents, so as to extract new knowledge from the data.