A Biologically Inspired Classifier

TL;DR

This paper extends a correlation-based distance measurement method to nonlinear functions inspired by biological problems, demonstrating robustness in detecting similarities despite translocations in data sequences.

Contribution

The paper introduces a nonlinear extension of a correlation-based classifier, inspired by biological microarray matching, and evaluates its robustness in similarity detection.

Findings

Method is robust against translocations in data sequences.

Nonlinear matching improves similarity detection accuracy.

Extension applicable to biological data analysis.

Abstract

We present a method for measuring the distance among records based on the correlations of data stored in the corresponding database entries. The original method (F. Bagnoli, A. Berrones and F. Franci. Physica A 332 (2004) 509-518) was formulated in the context of opinion formation. The opinions expressed over a set of topic originate a ``knowledge network'' among individuals, where two individuals are nearer the more similar their expressed opinions are. Assuming that individuals' opinions are stored in a database, the authors show that it is possible to anticipate an opinion using the correlations in the database. This corresponds to approximating the overlap between the tastes of two individuals with the correlations of their expressed opinions. In this paper we extend this model to nonlinear matching functions, inspired by biological problems such as microarray (probe-sample pairing). We investigate numerically the error between the correlation and the overlap matrix for eight sequences of reference with random probes. Results show that this method is particularly robust for detecting similarities in the presence of translocations.