High-throughput data analysis in behavior genetics

TL;DR

This paper presents robust computational methods for high-throughput analysis of behavioral data in genetics, addressing noise, outliers, and protocol deviations to improve accuracy in tracking and measuring rodent behavior.

Contribution

It introduces a combination of robust statistical techniques and algorithms tailored for analyzing large behavioral datasets in genetics research.

Findings

Effective estimation of mouse location, velocity, and acceleration.

Robust detection of arena boundary and center.

Automatic correction for protocol deviations.

Abstract

In recent years, a growing need has arisen in different fields for the development of computational systems for automated analysis of large amounts of data (high-throughput). Dealing with nonstandard noise structure and outliers, that could have been detected and corrected in manual analysis, must now be built into the system with the aid of robust methods. We discuss such problems and present insights and solutions in the context of behavior genetics, where data consists of a time series of locations of a mouse in a circular arena. In order to estimate the location, velocity and acceleration of the mouse, and identify stops, we use a nonstandard mix of robust and resistant methods: LOWESS and repeated running median. In addition, we argue that protection against small deviations from experimental protocols can be handled automatically using statistical methods. In our case, it is of biological interest to measure a rodent's distance from the arena's wall, but this measure is corrupted if the arena is not a perfect circle, as required in the protocol. The problem is addressed by estimating robustly the actual boundary of the arena and its center using a nonparametric regression quantile of the behavioral data, with the aid of a fast algorithm developed for that purpose.