A Classification of Bioinformatics Algorithms from the Viewpoint of Maximizing Expected Accuracy (MEA)

Abstract

Many estimation problems in bioinformatics are formulated as point estimation problems in a high-dimensional discrete space. In general, it is difficult to design reliable estimators for this type of problem, because the number of possible solutions is immense, which leads to an extremely low probability for every solution-even for the one with the highest probability. Therefore, maximum score and maximum likelihood estimators do not work well in this situation although they are widely employed in a number of applications. Maximizing expected accuracy (MEA) estimation, in which accuracy measures of the target problem and the entire distribution of solutions are considered, is a more successful approach. In this review, we provide an extensive discussion of algorithms and software based on MEA. We describe how a number of algorithms used in previous studies can be classified from the viewpoint of MEA. We believe that this review will be useful not only for users wishing to utilize software to solve the estimation problems appearing in this article, but also for developers wishing to design algorithms on the basis of MEA.

CentroidAlign-Web: A Fast and Accurate Multiple Aligner for Long Non-Coding RNAs

“…Note that CentroidAlign employs an estimator based on maximum expected accuracy (MEA), which has been successfully applied in much software in the field of bioinformatics; see the review by Hamada and Asai [24] for details. In CentroidAlign, the sum-of-pair scores (SPS) [25] is optimized for predicting multiple alignments of RNA sequences (cf.…”

“…( a ) The input is two RNA sequences, ( x; x ′ ), to be aligned; ( b-1 ) The exact algorithm of CentroidAlign considers a probability distribution of structural alignments between x and x ′ , which gives simultaneously the alignments between nucleotides and those between base-pairs (e.g., Sankoff model [4]); ( b-2 ) The exact case can be approximated by factorizing the distribution of structural alignments into (i) a distribution of secondary structures of x (e.g., the CONTRAfold [27] or McCaskill [22] models); (ii) a distribution of pairwise alignments between x and x ′ (e.g., the CONTRAlign model [23]); and (iii) a distribution of secondary structures of x ′ ; ( c ) By marginalization of the distribution(s) in (b), we obtain a distribution of alignments (*) in which the information about secondary structures is included; ( d ) The best multiple alignment is estimated based on maximizing expected accuracy (MEA) [24] in which the SPS scores of predicted alignments are optimized with respect to the distribution (*) of pairwise alignments given in (c). It should be emphasized that the computational cost of the exact algorithm is ≈ O ( L 6 ), while it is reduced to ≈ O ( L 3 ) in the approximate algorithm, where L is the (maximum) length of two input sequences.…”

Fighting against uncertainty: an essential issue in bioinformatics

“…where G(θ, y) is called the gain function, which returns the similarity between two solutions in Y . When the gain function is designed according to an accuracy or evaluation measure for the target problem, in which y and θ are considered as a prediction and reference, respectively, the estimator is often called a maximum expected accuracy (MEA) estimator [57,58,59] 6 . MEA estimators predict the solution by maximizing the expected accuracy when the solutions are distributed according to p(θ|D).…”

“…In addition to the above examples, many algorithms in bioinformatics can be classified, from the viewpoint of MEA/MEG, with respect to gain function and predictive space. See [59] for a review of MEA estimators.…”

Sequence annotation with HMMs: New problems and their complexity

“…Many other decoding criteria were proposed (Hamada and Asai, 2012). For example, we can assign labels to states of the HMM, and then search for the most probable sequence of labels instead of the most probable state path.…”