Predicting membrane protein types by the LLDA algorithm

Membrane proteins are generally classified into the following eight types: (1) type I transmembrane, (2) type II, (3) type III, (4) type IV, (5) multipass transmembrane, (6) lipid-chain-anchored membrane, (7) GPI-anchored membrane, and (8) peripheral membrane (K.C. Chou and H.B. Shen: BBRC, 2007, 360: 339-345). Knowing the type of an uncharacterized membrane protein often provides useful clues for finding its biological function and interaction process with other molecules in a biological system. With the explosion of protein sequences generated in the Post-Genomic Age, it is urgent to develop an automated method to deal with such a challenge. Recently, the PsePSSM (Pseudo Position-Specific Score Matrix) descriptor is proposed by Chou and Shen (Biochem. Biophys. Res. Comm. 2007, 360, 339-345) to represent a protein sample. The advantage of the PsePSSM descriptor is that it can combine the evolution information and sequence-correlated information. However, incorporating all these effects into a descriptor may cause the "high dimension disaster". To overcome such a problem, the fusion approach was adopted by Chou and Shen. Here, a completely different approach, the so-called LLDA (Local Linear Discriminant Analysis) is introduced to extract the key features from the high-dimensional PsePSSM space. The dimension-reduced descriptor vector thus obtained is a compact representation of the original high dimensional vector. Our jackknife and independent dataset test results indicate that it is very promising to use the LLDA approach to cope with complicated problems in biological systems, such as predicting the membrane protein type.