Chemometric Approaches for Systems Biology

Fortuny, Abel Folch

doi:10.4995/thesis/10251/77148

Cited by 2 publications

(3 citation statements)

References 291 publications

(465 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…With ideal, error-free data, the value of È is zero if the dot products x i Áx j exactly reproduce the correlation coefficients r ij . For a given case, the eigenvalues can be calculated after replacing ('imputing'; Karhunen, 2011;Folch-Fortuny et al, 2015) missing values in r = {r ij } by those computed, for example, from a least-squares solution. The minimum value of n can then be identified because it corresponds to the number of strong eigenvalues.…”

Section: Methods and Theorymentioning

confidence: 99%

Dissecting random and systematic differences between noisy composite data sets

Diederichs

2017

Acta Crystallogr D Struct Biol

View full text Add to dashboard Cite

Composite data sets measured on different objects are usually affected by random errors, but may also be influenced by systematic (genuine) differences in the objects themselves, or the experimental conditions. If the individual measurements forming each data set are quantitative and approximately normally distributed, a correlation coefficient is often used to compare data sets. However, the relations between data sets are not obvious from the matrix of pairwise correlations since the numerical value of the correlation coefficient is lowered by both random and systematic differences between the data sets. This work presents a multidimensional scaling analysis of the pairwise correlation coefficients which places data sets into a unit sphere within low-dimensional space, at a position given by their CC* values [as defined by Karplus & Diederichs (2012),Science,336, 1030–1033] in the radial direction and by their systematic differences in one or more angular directions. This dimensionality reduction can not only be used for classification purposes, but also to derive data-set relations on a continuous scale. Projecting the arrangement of data sets onto the subspace spanned by systematic differences (the surface of a unit sphere) allows, irrespective of the random-error levels, the identification of clusters of closely related data sets. The method gains power with increasing numbers of data sets. It is illustrated with an example from low signal-to-noise ratio image processing, and an application in macromolecular crystallography is shown, but the approach is completely general and thus should be widely applicable.

show abstract

Section: Methods and Theorymentioning

confidence: 99%

Dissecting random and systematic differences between noisy composite data sets

Diederichs

2017

Acta Crystallogr D Struct Biol

View full text Add to dashboard Cite

show abstract

“…This paper focuses on the generation of a PLS model based on its proven ability within industry to enable on-line predictions of off-line variables [47,48] and missing data. [18,19] Additionally two linear interpolation methods (linear and cubic splines) were applied to compare against the PLS model predictions.…”

Section: Wwwadvancedsciencenewscom Wwwbiotechnology-journalcommentioning

confidence: 99%

“…[9][10][11][12] The pioneering work by Wold et al (1987), [13] Nomikos and MacGregor (1994), [14] and Eriksson (1999) [15] has extended the fundamental principles of principal component analysis (PCA) and partial least squares (PLS) to enable direct comparison of batch-to-batch fermentation systems and the analysis of within batch variability. These techniques provide the necessary mathematical basis for on-line monitoring, [16] root cause analysis, [17] missing data algorithms, [18,19] and real-time control. [20][21][22] More recent advancements include the development of a MVDA tool kit to simplify the scale-up from ambr15 experiments to pilot-scale (300 L), resulting in shorter process development timelines and reduced costs.…”

mentioning

confidence: 99%

Multivariate Data Analysis Methodology to Solve Data Challenges Related to Scale‐Up Model Validation and Missing Data on a Micro‐Bioreactor System

Goldrick

Sandner

Cheeks³

et al. 2019

Biotechnology Journal

View full text Add to dashboard Cite

Multivariate data analysis (MVDA) is a highly valuable and significantly underutilized resource in biomanufacturing. It offers the opportunity to enhance understanding and leverage useful information from complex high‐dimensional data sets, recorded throughout all stages of therapeutic drug manufacture. To help standardize the application and promote this resource within the biopharmaceutical industry, this paper outlines a novel MVDA methodology describing the necessary steps for efficient and effective data analysis. The MVDA methodology is followed to solve two case studies: a “small data” and a “big data” challenge. In the “small data” example, a large‐scale data set is compared to data from a scale‐down model. This methodology enables a new quantitative metric for equivalence to be established by combining a two one‐sided test with principal component analysis. In the “big data” example, this methodology enables accurate predictions of critical missing data essential to a cloning study performed in the ambr15 system. These predictions are generated by exploiting the underlying relationship between the off‐line missing values and the on‐line measurements through the generation of a partial least squares model. In summary, the proposed MVDA methodology highlights the importance of data pre‐processing, restructuring, and visualization during data analytics to solve complex biopharmaceutical challenges.

show abstract

Chemometric Approaches for Systems Biology

Cited by 2 publications

References 291 publications

Dissecting random and systematic differences between noisy composite data sets

Dissecting random and systematic differences between noisy composite data sets

Multivariate Data Analysis Methodology to Solve Data Challenges Related to Scale‐Up Model Validation and Missing Data on a Micro‐Bioreactor System

Contact Info

Product

Resources

About