Multi-block methods for prediction and interpretation

Most of the disciplines within food science today produce large amounts of data. Data analytical tools to predict, interpret and understand these data are essential for achieving results and gaining new knowledge. In this program, we have developed methods for so-called "multiblock" data analysis, i.e. finding relations between many data tables. The program has worked closely with the users of these methods, which include scientists working with spectroscopy, microbiology, functional genomics, sensory and consumer research. Most of the methodology is generic and applicable to problems in all disciplines. A collection of methods for making predictions, classifications and interpretation based on several data blocks has been developed. The new methods are called SO-PLS ("Sequential Orthogonalized PLS"), PO-PLS ("Parallel Orthogonalized PLS"), ROSA ("Response-Oriented Sequential Alternation") and SMI ("Similarity Matrix Index"). Several extensions of the methods have also been developed, such as tools to identify important variables and a variation of SO-PLS that can be used for multidimensional data blocks. Within spectroscopy, the methods have for instance been used to analyze spectra from several different instruments simultaneously (e.g. near-infrared, infrared, Raman and fluorescence spectroscopy). The objective of these analysis have been to understand the information contained in each of the data blocks and exploit complementary information for better predictions of various chemical components and quality characteristics. A major focus over the past year has been to analyze the relationship between IR spectra and size exclusion chromatography, and it has been shown that IR spectroscopy can be used to monitor and optimize industrial processes for enzymatic protein hydrolysis. In microbiology, the methods have been used to analyze the intestinal microbiota of both humans and animals, and understand how different diets affect the microbiota. The methods have also been used to model the growth of pathogenic and spoilage bacteria in food, with regard to on food safety and shelf life. An analysis of bacterial communities in fish tanks has also been conducted, in order to examine how the bacteria evolve at different salinity levels. In addition, a study to compare different statistical methods for analyzing temporal effects on microbiota has been initiated. Within functional genomics (including proteomics), methods for analyzing data from wheat, fish, Lactobacillus, Arabidopsis and humans have been developed and applied. For wheat, genes that break the negative correlation between yield and protein content have been identified. For Lactobacillus, we have studied the basic mechanisms for how different strains adapt to variations in nutrients. In the human study, the results suggest that vitamin D deficiency may be associated with multiple sclerosis (MS). Within sensory and consumer research, focus has been on development and comparison of methods for analyzing data from rapid sensory profiling techniques such as "projective mapping" and CATA ("check-all-that-apply"), as well as analysis of data from multiblock L-structures. The latter typically involves situations where the data blocks correspond to consumer preferences, information about the consumers (gender, age attitudes etc.) and information about the samples analyzed. Methodology for analyzing individual differences between consumers and for clustering different consumer segments have also been developed. This program has experiences important synergies between application areas. For instance, a well-established method of clustering consumer segments has been used in microbiology, to identify groups of bacteria that react differently to a diet or treatment. A publication that formalizes the concepts of "common" and "unique" information in multi-block analysis has also been published, in which the relevance to spectroscopy, sensory science and metabolomics is shown in concrete case studies. This work is conducted in collaboration with scientist at the University of Copenhagen and the University of Amsterdam, and allows for better understanding of the difference between the established methods in the field. Synergies are also highlighted in several scientific publications, comprising practical examples of the same methods from both spectroscopy and sensors sensory science.

I matindustrien og matforskningen er det i dag et stort behov for å forstå relasjoner mellom flere datasett som samles inn. Dette henger sammen med kompleksiteten innen det enkelte fagfelt og med det faktum at for å løse de større problemer i sektoren har man behov for kommunikasjon på tvers av tradisjonelle faggrenser. Eksempler der slik forståelse av flere datasett er svært viktig er innen forbrukerforskning, prosessmodellering, mikrobiologi, emballering, funksjonell genomikk samt mat og helseforskning (se f.eks. Horizon 2020 dokumentet). Alle disse områdene er sentrale for norsk matindustri (se NTP, Food for Life) og for Nofima sin prosjektportefølje (FFL baserte prosjekt og andre). Mer konkret kan nevnes produktutvikling i industrien hvor man har beho v for informasjon om kjemisk sammensetning av produkter, sensoriske egenskaper og forbrukerpreferanser samt kunnskap om holdbarhet og helseeffekter. Alle de nevnte fagområdene produserer store datamengder og for å få til en god samhandling er det en nødve ndig forutsetning å ha gode statistiske verktøy til å prediktere, tolke og forstå relasjonen mellom datasettene som lages. Det finnes allerede metoder for dette, men metodeutviklingen ligger langt bak de aktuelle behov innen både forskningen og for indust rien. Dette strategiske proprammet vil ha fokus på utvikling av relevante og brukervennlige multivariate statistiske og kjemometriske metoder for analyse og tolkning av relasjoner mellom flere store datasett. Programmet vil arbeide i tett relasjon med fag områdene nevnt over for metodeutvikling for løsning av konkrete problem. Programmet vil således bidra sterkt til tverrfaglighet og kommunikasjon mellom en serie andre aktiviteter av stor relevans for matbransjen. Metodene som utvikles er generiske og vil ha potensielle anvendelser også i andre industrier og innen f.eks. medisin.