The problem of identifying genes in genomic DNA sequences by computational methods has attracted considerable research attention in recent years. From one point of view, the problem is closely

We present a statistical model of genes in DNA. A Generalized Hidden Markov Model (GHMM) provides the framework for describing the grammar of a legal parse of a DNA sequence (Stormo & Haussler 1994). Probabilities are assigned to transitions between states in the GHMM and to the generation of each nucleotide base given a particular state. Machine learning techniques are applied to optimize these probabilities using a standardized training set. Given a new candidate sequence, the best parse is deduced from the model using a dynamic programming algorithm to identify the path through the model with maximum probability. The GHMM is flexible and modular, so new sensors and additional states can be inserted easily. In addition, it provides simple solutions for integrating cardinality constraints, reading frame constraints, "indels", and homology searching. The description and results of an implementation of such a gene-finding model, called Genie, is presented. The exon sensor is a codon fre...

We present an improved splice site predictor for the genefinding program Genie. Genie is based on a generalized Hidden Markov Model (GHMM) that describes the grammar of a legal parse of a multi-exon gene in a DNA sequence. In Genie, probabilities are estimated for gene features by using dynamic programming to combine information from multiple content and signal sensors, including sensors that integrate matches to homologous sequences from a database. One of the hardest problems in genefinding is to determine the complete gene structure correctly. The splice site sensors are the key signal sensors that address this problem. We replaced the existing splice site sensors in Genie with two novel neural networks based on dinucleotide frequencies. Using these novel sensors, Genie shows significant improvements in the sensitivity and specificity of gene structure identification. Experimental results in tests using a standard set of annotated genes showed that Genie identified 86% of coding nuc...

We consider the problem of parsing a sequence into different classes of subsequences. Two common examples are finding the exons and introns in genomic sequences and identifying the secondary structure domains of protein sequences. In each case there are various types of evidence that are relevant to the classification, but none are completely reliable, so we expect some weighted average of all the evidence to provide improved classifications. For example, in the problem of identifying coding regions in genomic DNA, the combined use of evidence such as codon bias and splice junction patterns can give more reliable predictions than either type of evidence alone. We show three main results: 1. For a given weighting of the evidence a dynamic programming algorithm returns the optimal parse and any number of sub-optimal parses. 2. For a given weighting of the evidence a dynamic programming algorithm determines the probability of the optimal parse and any number of sub-optimal parses under a ...

We introduce a new system, called shortHMM, for predicting exons, which predicts individual exons using two related genomes. In this system, we build a hidden semi-Markov model to identify exons. In the hidden Markov model, we propose joint probability models of nucleotides in introns, splice sites, 5 ′ UTR, 3 ′ UTR and intergenic regions by exploiting the homology between related genomes. In order to reduce the false positive rate of the hidden Markov model, we develop a screening process which is able to identify intergenic regions. We then build a classifier by combining the statistics from the hidden Markov model and the screening process. We implement shortHMM on human-mouse sequence alignments. Compared to TWINSCAN and SLAM, shortHMM is substantially more powerful in identifying AT-rich RefSeq exons (8 % more AT-rich RefSeq exons were predicted), as well as slightly more powerful in identifying RefSeq exons (3%-10 % more RefSeq exons were predicted), at a similar or lower false positive rate, with less computing time and with less memory usage. Last, shortHMM is also capable of finding new potential exons.