The Gene Identification Problem from a Hypotheses Test Perspective

Many gene identification methods assign scores to gene elementsas a previous step to their assembly in predicted genes. The scoring system is often based on log likelihood ratios (LLRs) whose meaning is somehow different from the usual likelihood ratio tests that appear in many statistical problems. In the first part of this work we have tried to give an interpretation of the statistical meaning of LLRs based scoring systems. We have developed several tests of significance for the scores: the " Sum-of-Scores test " (SSt), based on the straightforward score obtained by the programs, the " Intersection-Union test " (IUt) based on a multiple hypothesis testing interpretation of an exon's score and several meta-analytical approaches which combines p-values corresponding to the exon's parts. We have performed simulation studies to analyze the performance of these tests. Whereas SSt and IUt tests are appealing from the statistical point of view the meta-analytic approach has proved to have a much better sensitivity and specificity which suggests they may be incorporated in actual gene prediction methods as a com-plimentary " probabilistic " score. To approximate the distribution of the tests under the null hypothesis (of non-exonicity) two approaches are possible: to resample from real " non-coding-no sites " data or to simulate them using the scoring model and the estimated probability matrices. Although both approaches should be, in principle, equivalent it is interesting to compare them in order to check the validity of the scoring system used by GeneId. A comparison between both approaches has been performed in the second part of this work using Monte Carlo simulation. The main conclusion is that using probability matrices and the scoring system is essentially equivalent to resampling only in the case of sites (start, acceptor donor or stop). In the case of coding potential they are not always, equivalent and in general resampling is preferred to simulation. In order to perform the analysis and tests described a java program has been implemented. It can be used in several ways: (1) It can take GeneId output and use it to perform some or all tests of exonicity. (2) It can be used to generate samples of true or false exons (partial or complete) that can be used as input for other studies such as testing gene prediction programs .