Contribution of multiple factor analysis to sensory data study

Abstract

<p style="text-align: justify;">Multiple Factor Analysis (MFA) deals with data in which a set of individuals is described by several sets of variables. Such data are frequently encountered in sensory analysis, for example whcn we wartt to compare panels, or to point out relationships between sensory data and chemical data. We present an application of MFA to data in which 50 sparkling wines (including 26 champagnes) are evaluated by 32 assessors (amateurs and oenologists) through 24 descriptors. Here, wines play the role of individuals ; the variables are the 32 x 24 descriptors ; one group gathers descriptors associated to a single assessor.</p><p style="text-align: justify;">In this example, some questions are particularly important.</p><p style="text-align: justify;">What are the main factors in the perception of these wines ? Are descriptors correlated ? Were champagnes perceived different than the other wines ? Do amateurs perceived these wines as oenologists ? Are chemical data correlated to sensory data ?</p><p style="text-align: justify;">This application shows the interest of MFA, which provides firstly classical results of factor analysis. Thus, graphical displays of wines and of descriptors point out a clear opposition between the champagnes and the other wines. Champagnes were perceived more sparkling (this result is interesting because effervescence is subjected to a glass effect which usually masks differences between wines), with a stronger taste and aroma of old wines. From a chemical point of view, champagnes have a high measured effervescence and a low level of S02.</p><p style="text-align: justify;">MFA provides also results specifie to such multiple tables :</p><p style="text-align: justify;">- graphical displays of variables groups ; here a group corresponds to a judge (each one contains the descriptors used by one judge) that is to say to the wines configuration associated to one judge. On this graphie, judges who globally perceived the wines in the same inanner are close one to the other. In this application, surprisingly, there is no clear distinction between amateurs and oenologist.</p><p style="text-align: justify;">- graphical displays of wines according to each judge. The 32 configurations of wines (each one for a singlejudge) are superimposed, as in procrustes analysis (the principles of the two methods are different but, from the point of view of this graphie, they are similar). This study shows wines perceived quite in the same manner by the different judges, and wines which are subject of various judgements.</p><p style="text-align: justify;">- a set of canonical correlation coefficients : here they indicate that the first factor of MFA, which opposes champagnes to the other wines, is common to quite all groups (that is to say to all judges).</p><p style="text-align: justify;">All these results derive from a single analysis. Thus it is possible to study, in a unique framework, all the aspects of wines variability and judges variability.</p>

Multi-omics integration—a comparison of unsupervised clustering methodologies

“…Following covariance optimization between the global score derived from the concatenated matrix and single omics scores, this method was applied to mRNA, miRNA and proteomics data, and succeeded in distinguishing profiles from melanoma, leukemia and central nervous system cell lines [45]. Furthermore, multiple factor analysis (MFA) [43,46] is a concatenation-based method whose strategy is instead based on the principal component analysis (PCA) of the concatenated matrix. MFA was applied in [43] to copy-number measurements and gene expression from a glioma data set to study differences between different tumor subtypes.…”

High FLT3 expression increases immune‐cell infiltration in the tumor microenvironment and correlates with prolonged disease‐free survival in patients with non‐small cell lung cancer

“…As a surrogate for sDCs and NK cells abundance, expression of previously described signature genes of sDCs ( KIT , CCR7 , BATF3 , FLT3 (excluded from MCPcounter analysis), ZBTB46 , IRF8 , BTLA , MYCL1 ) and NK ( GNLY , KLRC3 , FLT3LG KLRD1 , KLRF1 , NCR1 ) has been adopted [ 9 , 26 , 27 ]. Spatial partitioning of LUSC and LUAD patients regarding their clinical characteristics was performed based on models of FLT3 ‐low/ FLT3 ‐high and T‐cell exhaustion marker gene expression ( PDCD1 (PD1), CD274 (PDL1), PDCD1LG2 (PDL2), CTLA4 , LAG3 , HAVCR2 (TIM3), GZMB , BTLA , CD160 , CD244 (2B4), TIGIT ) through the Multiple Factor Analysis (MFA) being an extension of the Principal Component Analysis (PCA) allowing to mix variables of different types [ 28 , 29 , 30 , 31 ].…”