Similarity searching is one of the most widely used techniques for virtual screening in large-scale drug discovery programmes. As one of the principal components, molecular descriptors play a crucial role in similarity searching. However, there is not yet one individual set of descriptors describing compounds perfectly so far. Recently, some research groups carried out similarity searching by fusing different fingerprints types. As these descriptors are all based on the structure of compounds, their fusions are more likely to cause redundancy far from describing compounds more comprehensively. Here, the Gene Ontology (GO) fingerprint is presented as a novo type of descriptor on the basis of GO terms and used to describe compounds together with structure fingerprint under the philosophy principle of describing substance from both nature and extend. The application of GO fingerprint reduces the dimensions of microarray data and avoids its problems such as high dimensions, strong correlation and so on. It also shortens the distance between the descriptors and the biological activities of compounds. Comparing of the similarity searching results derived from the methods using the structure fingerprint or the GO fingerprint only, the integrated method using both types shows much better description ability in two aspects. First, it can give a higher score to the compounds similar to the query in both structure and bioactivity. Second, it can effectively get rid of the compounds special in one side. As a conclusion, we propose a novo way for fast, effective, high-throughput drug discovery. It is expected to improve the results of virtual screening and accelerate the process of drug discovery and reuse of old drugs.

Key words: structure fingerprint, function fingerprint, gene ontology, drug screening, molecular descriptors