PRED-TMBB

**PRED-TMBB2: An Updated and Improved Version of the Pred-TMBB Method for Predicting Outer Membrane Proteins

** The PRED-TMBB method is a software tool based on Hidden Markov Models that predicts the topology of beta-barrel outer membrane proteins and distinguishes them from water-soluble ones. The new version, PRED-TMBB2, has undergone significant improvements that enhance its performance. One of the enhancements is including a properly defined end state that allows for better modeling of the beta-barrel domain. In addition, different emission probabilities for the adjacent residues in strands are used to incorporate knowledge about the asymmetric amino acid distribution occurring there.

Furthermore, the training of PRED-TMBB2 was optimized using newly developed algorithms in order to improve the accuracy of the labels of the training sequences. Additionally, the method is retrained on a larger, non-redundant dataset that includes recently solved structures. This ensures that the tool can accurately predict the topology of the beta-barrel outer membrane proteins. A newly developed decoding method was also added to the already available options. Finally, the method now allows the incorporation of evolutionary information in multiple sequence alignments, further enhancing its accuracy.

A strict cross-validation procedure results show that PRED-TMBB2 with homology information performs significantly better than other available prediction methods. It yields 76% correct topology predictions, outperforming the best available predictor by 7%, with an overall Segment Overlap Score (SOV) of 0.9. Moreover, PRED-TMBB2 can detect beta-barrel proteins with a Matthews Correlation Coefficient (MCC) value of 0.92 using the query sequence as input.