This Project presents a methodology to select between Geometric and Bio-inspired distance metrics in a semi-supervised classifier using Support Vector Machine (SVM) to classify protein sequences from land plants (Embryophyta dataset). First, a kernel matrix was built in a process of extraction and feature selection, on the other hand, another matrix was built to Euclidean, Mahalanobis, Mismatch and Gappy distances. Both matrices were used in the Neighborhood kernel algorithm to obtain a semi-supervised matrix to an optimized SVM classifier using PSO and W-SVM. The prediction model was evaluated calculating a confusion matrix between training data and test data, with partitions from cross-validation method; after was calculated a geometric mean with the sensitivity and specificity. The results show that the methodology presented is efficient to select the best metric according to the molecular function. The Euclidean metric was selected as the best one for seven functions, with score from 49.94% to 74.3%. Mismatch was selected for three functions, with score from 51.63% to 80.78%, and Gappy was selected for four functions, with score from 43.11% to 68.5%. On the other hand, it is important to stand out that this work allowed to create a new research line in Bioinformatic algorithm in the ITM, in addition, this one derived four Degree works in Engineering and two new students of Maestria en Automatizacion y Control industrial