Partial least squares correspondence analysis: A framework to simultaneously analyze behavioral and genetic data.

Abstract

For nearly a century, detecting the genetic contributions to cognitive and behavioral phenomena has been a core interest for psychological research. Recently, this interest has been reinvigorated by the availability of genotyping technologies (e.g., microarrays) that provide new genetic data, such as single nucleotide polymorphisms (SNPs). These SNPs-which represent pairs of nucleotide letters (e.g., AA, AG, or GG) found at specific positions on human chromosomes-are best considered as categorical variables, but this coding scheme can make difficult the multivariate analysis of their relationships with behavioral measurements, because most multivariate techniques developed for the analysis between sets of variables are designed for quantitative variables. To palliate this problem, we present a generalization of partial least squares-a technique used to extract the information common to 2 different data tables measured on the same observations-called partial least squares correspondence analysis-that is specifically tailored for the analysis of categorical and mixed ("heterogeneous") data types. Here, we formally define and illustrate-in a tutorial format-how partial least squares correspondence analysis extends to various types of data and design problems that are particularly relevant for psychological research that include genetic data. We illustrate partial least squares correspondence analysis with genetic, behavioral, and neuroimaging data from the Alzheimer's Disease Neuroimaging Initiative. R code is available on the Comprehensive R Archive Network and via the authors' websites. (PsycINFO Database Record

Multivariate genotypic analyses that identify specific genotypes to characterize disease and control groups in ADNI

“…In general, PLS-CA is the generalization of partial least squares-a family of techniques that analyze the information common to two data tables [50]-that can be used with virtually any data type [49,60]. We used discriminant and seed PLS-CA in our studies here.…”

“…Because we treat our data categorically we required particular multivariate techniques designed specifically for categorical data. We used multiple correspondence analysis (MCA) and two forms of partial least squares-correspondence analysis (PLS-CA): discriminant PLS-CA and seed PLS-CA [49]. We briefly describe these techniques here but also provide more details where necessary.…”

“…To achieve this, we used a multivariate technique designed specifically for categorical data called partial least squares-correspondence analysis (PLS-CA, [49]). PLS-CA is a generalization of two different families of techniques: partial least squares (PLS; as in neuroimaging: [50,51]), and correspondence analysis (CA; [52][53][54][55]).…”

“…PLS-CA analyzes the relationships between two tables of categorical or mixed data. In our study, we used two specific forms of PLS-CA called discriminant PLS-CA (a.k.a., mean-centered PLS-CA or discriminant correspondence analysis; [56]) and "seed" PLS-CA (see Methods and [49] for more details).…”

“…SNPs were then recoded into a disjunctive format as seen in Table 2 (for more details on disjunctive format and PLS-CA see [49]). Some genotypes could occur below a reasonable threshold (e.g., 5%) and therefore we performed an additional preprocessing step: genotypes that fell below a 5% threshold were combined with another genotype.…”