Computational Analysis of Neonatal Mouse Ultrasonic Vocalization

Abstract

Neonatal vocalization is structurally altered in mouse models of autism spectrum disorder (ASD). Our published data showed that pup vocalization, under conditions of maternal separation, contains sequences whose alterations in a genetic mouse model of ASD impair social communication between pups and mothers. We describe details of a method which reveals the statistical structure of call sequences that are functionally critical for optimal maternal care. Entropy analysis determines the degree of non-random call sequencing. A Markov model determines the actual call sequences used by pups. Sparse partial least squares discriminant analysis (sPLS-DA) identifies call sequences that differentiate groups and reveals the degrees of individual variability in call sequences between groups. These three sets of analyses can be used to identify the otherwise hidden call structure that is altered in mouse models of developmental neuropsychiatric disorders, including not only autism but also schizophrenia. © 2018 by John Wiley & Sons, Inc.

Keywords: Markov model; USV; autism spectrum disorder; entropy; mouse models; sPLS-DA; schizophrenia.

Figure 1

A representative spectrogram. Taken from Takahashi et al., (2016).

Figure 2

An example to determine the cut-off interval beyond which to judge two calls which are not parts of a sequence. Blue solid and broken arrows indicate the cross points of wild-type and heterozygous groups, respectively (Takahashi et al., 2016).

Figure 3

Entropy plot. Red boxplots show an example of the range of entropy scores for Tbx1 heterozygous (HT) mice at each level in the entropy analysis; blue boxplots show the corresponding scores for wild-type (WT) mice.

Figure 4

Connections and their directions between two calls identified using the Markov model. Thickness of arrows and size of call circles represent the relative proportion of a connection and call numbers (Takahashi et al., 2016).

Figure 5

Scores plot. The scores plot depicts the samples, colored according to class (genotype), projected into the space defined by the first two sPLS-DA model components. Axis labels show the level of variance explained (expl. var) by each component. sPLS-DA, sparse partial least squares discriminant analysis.

Figure 6

Loading plot. The loading plot shows the strength of the association for each of the selected variables on a particular component (Comp) with the class labels (genotypes).