Objective      To establish a new machine learning model

and to predict outer membrane proteins (OMPs) from various protein datasets and genomic protein sequences using the methods with better accuracy than existing methods.        Methods      For protein sequences

combined features were constructed by encoding based on grouped weights and computing amino acid compositions

and the F-score technique was used to obtain backward selection subsets

which were then inputted to a classifier based on support vector machine algorithms (SVM). The classification model was tested on a dataset including 1 087 protein sequences for evaluation of sensitivity

specificity and overall accuracy

and was also trained and performed to predict OMPs in multiple bacterial genomes.       Results        Through combing the new sequence encoding methods with a SVM-based classifier

our model discriminated OMPs from α-helical transmembrane proteins

globular proteins and non-OMPs with cross-validated accuracy of 94.7%

96.4% and 94.6%

respectively

and the accuracy severally increased to 95.7%

96.9% and 95.9% after the feature selection. The whole-genomic prediction also fit the known facts reasonably.       Conclusion      The presented model shows better accuracy than other methods based on sequence features in the literature

and can be used for screening OMPs in genomic sequences.