This thesis develops methods to classifiy the substrates transported across a membrane by a given transmembrane protein. Our methods use tools that predict specificity determining sites (SDS) after computing a multiple sequence alignment (MSA), and then building a profile Hidden Markov Model (HMM) using HMMER. In bioinformatics, HMMER is a set of widely used applications for sequence analysis based on profile HMM. Specificity determining sites (SDS) are the key positions in a protein sequence that play a crucial role in functional variation within the protein family during the course of evolution.

We have established a classification pipeline which integrated the steps of data processing, model building and model evaluation. The pipeline contains similarity search, multiple sequence alignment, specificity determining site prediction and construction of a profile Hidden Markov Model.

We did comprehensive testing and analysis of different combinations of MSA and SDS tools in our pipeline. The best performing combination was MUSCLE with Xdet, and the performance analysis showed that the overall average Matthews Correlation Coefficient (MCC) across the seven substrate classes of the dataset was 0.71, which outperforms the state-of-the-art.