Kemometriska strategier för guidad multi-modellanalys

Understanding biological processes is inherently complex. The cellular machinery andbiochemical pathways present significant challenges in scientific research. Advances indata collection, such as high-content imaging and omics technologies, have enableddeeper insights, but extracting meaningful conclusions from these complicateddatasets remains a challenge. In this thesis, the focus has been on developingchemometric strategies and supervised modelling approaches to improve datainterpretation, aiming to aid scientists in drawing conclusions from their data.In Paper I, we show that cell imaging data, combined with chemometric tools, caneffectively characterize treatment effects, leading to the development of a metric calledEquivalence (Eq.) scores. This work raised two main questions: Are fluorescent labelsnecessary for meaningful characterization? Can living cells, imaged over time, providedeeper insights? In Paper III, we address these questions by investigating anapproach based on label-free live-cell imaging data where we extended the Eq. scoresto time series data. We demonstrate that time-dependent analysis reveals both earlyand late cellular responses and improves the prediction of drug mechanisms.In Paper II, we address challenges arising when Orthogonal Projections to LatentStructures-Discriminant Analysis (OPLS-DA) models are used to analyse severalclasses, such as subtypes of diseases or different treatments. We introduce OPLSHierarchicalDiscriminant Analysis (OPLS-HDA), a method that integrateshierarchical clustering analysis (HCA) with two-class OPLS-DA models to create anOPLS-based decision tree. We demonstrated that OPLS-HDA is a strong classifiercompared to eight other established methods while maintaining interpretability.Additionally, we provide Python scripts that are integrated with SIMCA®, offering auser-friendly interface for broader accessibility.Extracting reliable insights from complex data requires intentional and structuredapproaches. This work highlights the benefits of modular and interpretable modellingsolutions, ensuring that results are both understandable and trustworthy. By breakingdown complex analytical challenges and building tools that enhance interpretability,this work contributes to the broader goal of accelerating data-driven discoveries in lifesciences.