Sequence Retriever for Known, Discovered, and User-Specified Molecular Fragments

Abstract

Typically, biological and chemical data are sequential, for example, as in genomic sequences or as in diverse chemical formats, such as InChI or SMILES. That poses a major problem for computational analysis, since the majority of the methods for data mining and prediction were developed to work on feature vectors. To address this challenge, a functionality of a Statistical Adapter has been proposed recently. It automatically converts parsable sequential input into feature vectors. During the conversion, insights are gained into the problem via finding regions of interest in the sequence and the level of abstraction for their representation, and the feature vectors are filled with the counts of interesting sequence fragments, – finally, making it possible to benefit from powerful vector-based methods. For this submission, the Sequence Retriever has been added to the Adapter. While the Adapter performs the conversion: sequence → vector with the counts of interesting molecular fragments, the Retriever performs the mapping: molecular fragment → sequences from the database that contain this fragment.