Palavras-chave: modelos probabilísticos, predição de genes, cadeia de markov oculta generalizada, aprendizado de máquina, bioinformática.iii iv Abstract Lopes, Bruno Tenório da Silveira Ab initio gene prediction combined with alignment information. 2019. 70 p. DissertationIn Bioinformatics, the field of computational prediction of protein-coding genes is one of the most challenging and did not have many advances in the last decade. There are two main groups of methods for predicting genes: ab initio methods and extrinsic methods. The ab initio or intrinsic group includes the programs that perform the prediction using only the target sequence as input.This group focuses on the search for gene structures based on biological signals and preserved portions of the sequence. The other group, called extrinsic, consists of the programs that require other (reference) sequences in addition to the target sequence to perform the prediction by aligning the target sequence against reference sequences. There are also prediction approaches that attempt to join the two prediction methods, called the hybrid, incorporating alignments to increase the precision of the ab initio predictors. In this dissertation we developed an extension of the ToPS computational framework to implement two hybrid prediction techniques and assess their benefits and relative merits. The results obtained show a clear benefit from including genome alignments in the prediction and the pros and cons of using transcript mapping. Additionally, we have devised a generic model to include probabilistic extraneous information into a gene predictor. This model is implemented in ToPS and can be used to further develop gene prediction strategies.