Advances in Hidden Markov Models for Sequence Annotation

Finding genes in Schistosoma japonicum: annotating novel genomes with help of extrinsic evidence

“…The super advisor probability distribution is then estimated by a linear combination of these simplified advisor predictions. We have previously tested this method and found its performance satisfactory and robust to changes of parameters of individual advisors (12). In the final step, the super advisor scores are combined with the probabilistic distribution defined by the HMM (5) and the modified most probable annotation of the sequence is predicted by a modified Viterbi algorithm.…”

Theory of Segmentation

“…In statistics, this type of estimation is known as marginal posterior mode (Winkler, 2003) or maximum posterior marginals (Rue, 1995) (MPM) estimation. In computational biology, this is also known as the posterior decoding (PD) (Brejová et al, 2008). In the wider context of biological applications of discrete high-dimensional probability models this has also been called "consensus estimation", and in the absence of constraints, "centroid estimation" (Carvalho & Lawrence, 2008).…”

“…codon, intron, intergenic). In effect, this would realize the notion of path similarity by mapping many "similar" state paths to a single label path or annotation (Brejová et al, 2008;Fariselli et al, 2005;Käll et al, 2005;Krogh, 1997). However, this leads to the problem of multiple paths, which in practically important HMMs renders the dynamic programming approach of the Viterbi algorithm NP-hard (Brejová et al, 2007;Brown & Truszkowski, 2010).…”

“…Unlike the Viterbi classifier, PMAP handles annotations as easily as it does state paths, including the enforcement of the positivity constraint (Käll et al, 2005). A number of heuristic approaches are also known to alleviate these problems, but none appears to be fully satisfactory (Brejová et al, 2008). Note that mapping optimal state paths to the corresponding annotations need not lead to optimal annotations and can give poor results (Brejová et al, 2007).…”

StochHMM: a flexible hidden Markov model tool and C++ library

“…Since the first application of hidden Markov models (HMMs) to biological sequences in the 1980s, they have become a fundamental tool in bioinformatics. This is because of their robust statistical foundation, conceptual simplicity and malleability that allow researchers to adapt them to fit diverse classification problems (Brejova´and Brown, 2008). Many HMM-based programs have adapted the basic HMM framework to solve unique biological problems, such as development of generalized HMM (GHMM) to simplify state durations (Kulp et al, 1996), stochastic sampling to predict alternative splicing (Cawley and Pachter, 2003;Stanke et al, 2006a) and integration of additional data sources into the decoding algorithms to improve gene predictions (Stanke et al, 2006b).…”