A Paradigm Shift in Biological Computing - A Survey on the Development of Bioinformatics Algorithms in Taiwan

In this survey we focus on bioinformatics literature involving computational algorithms from authors in Taiwan. In a typical scenario of discrete algorithm design we have a single objective to optimize or a certain property to recognize. One can expect either to give a polynomial time algorithm for the problem or to show that it is NP-complete/hard. However, biological experiments are never flawless. The data involved in biocomputing algorithms is always noisy. Thus, the traditional rigid discrete algorithm does not seem to produce a viable solution for biologists. For an optimization problem it is sometimes possible to ask how close one can approximate to an optimal solution. However, it is more troublesome for a recognition problem given the noisy data. One can certainly modify the noisy version to an optimization problem by asking what is the minimum number of modifications (e.g. deletions, insertions or substitutions) to transform the noisy data into one that conforms to the prescribed property. But such a purely combinatorial approach usually suffers from the following two unpleasant phenomena: 1. The problem of finding this least number would likely become NP-hard; 2. Even if one can find the best modification in this sense, the data in the resultant configuration might not make any biological sense. There are many other complications to a biological computation problem such as multiple objective optimization with dependent objective functions, heterogeneous constraint types such as subset selection constraints, numerical weighting constraints, plus local noises that are less amenable to a global optimization approach. To resolve this, the key is to employ as much problem structures (both local and global) and features as possible in the algorithm design. Therefore, data dependent algorithms such as natural language and machine learning techniques are becoming increasingly more important. We will make an overview in the following areas: sequence analysis, sequence/structure alignment, protein structure prediction, phylogenetic analysis, motif finding, gene finding/annotation, protein structure prediction, protein structure identification and literature mining.