Despite many advances in the generation of high producing recombinant mammalian cell lines over the last few decades, cell line selection and development is often slowed by the inability to predict a cell line ’ s phenotypic characteristics (e.g. growth or recombinant protein productivity) at larger scale (large volume bioreactors) using data from early cell line construction at small culture scale. Here we describe the development of an intact cell MALDI-ToF mass spectrometry fingerprinting method for mammalian cells early in the cell line construction process whereby the resulting mass spectrometry data is used to predict the phenotype of mammalian cell lines at larger culture scale using a Partial Least Squares Discriminant Analysis (PLS-DA) model. Using MALDI-ToF mass spectrometry, a library of mass spectrometry fingerprints was generated for individual cell lines at the 96 deep well plate stage of cell line development. The growth and productivity of these cell lines were evaluated in a 10 L bioreactor model of Lonza’s large-scale (up to 20,000 L) fed-batch cell culture processes.
As the sample size gets larger, the misclassification rate becomes smaller for the PLS-DA. On the other hand, PCA+LDA are inconsistent. In conclusion, the entire results revealed that PLS-DA highly recommend for a large sample size than PCA+LDA in dimension reduction and classification. PLS-DA can be considered to have a good and reliable technique to be used when dealing with large datasets for classification task. The future work might include the real dataset to make comparison between the performance of PCA+LDA and PLS-DA on large variable dataset.
The ATR-FTIR spectroscopy-based diagnostic dem- onstrated excellent sensitivity and specificity against the PCR gold standard. The support vector machine learning algorithm performed better than the PLS-DA modelling approaches (AUROC = 0.98 vs 0.93). Empirical testing of modelling performance based on sample numbers (Addi- tional file 9: Fig. S6) indicated that sensitivity and speci- ficity are likely to improve further with calibration set numbers up to n = 500. When applying the SVM classifi- cation there were three false negatives and two false posi- tives. One case of misclassification could be accounted for; a false-positives classified positive by both micros- copy and RDT (Pan). It appears this patient was infected by a Plasmodium species other than P. falciparum or P. vivax, not detected by RDT or by the PCR primers employed.
We then tested whether plaque and calculus from a tooth affected by either caries or periodontal disease were equally distinct. A PCA plot indicated that the plaque and calculus samples were clustered mostly dis- tinctly but with a slight overlap between plaque and modern calculus (Fig. 2d), but group separation was not significant by adonis test. Fourteen species were more abundant in diseased tooth site calculus than healthy site plaque, and 13 were more abundant in healthy site plaque (q ≤ 0.05, effect size ≥ 1) (Fig. 2e). Of the 14 species with greater abundance in disease site calculus, 11 are also more abundant in healthy site calculus, but other species including Porphyromonas gingivalis, Treponema denticola, and Filifactor alocis, all of which are strongly associated with periodontal disease site plaque [17, 24], are significantly more abundant only when comparing disease site calculus to healthy site plaque. Sparse PLS-DA again demonstrated that our samples are sufficiently informative to classify healthy site plaque and disease site calculus based on microbial profile (Fig. 2f ), which was confirmed by the low classifi- cation balanced error rate (BER) (< 0.0008, Additional file 2: Table S3), and the species that contribute most to the classification are differentially abundant between the two groups (Additional file 1: Figure S4B). We repeated these analyses comparing healthy site plaque profiles to calculus only affected by periodontal disease (i.e., exclud- ing caries) and found nearly identical trends (Additional file 1: Figure S5A-D).
Partial least squares discriminant analysis (PLS- DA) is a discriminant analysis method based on partial least-squares regression. It has been widely used in developing multivariate classification models based on spectroscopic measurements. Smart PLS-DA (sPLS-DA) was used to classify the three types of endometrial tissue samples based on the preprocessed NIR data. It is an approach that automatically selects the optimal number of latent variables based on the average minimal prediction obtained from an internal bootstrap Latin partition (BLP) of the calibration set. sPLS-DA classification accuracies were used for evaluating and comparing the spectral preprocessing methods. 3. EXPERIMENTAL
not include a drug eﬀect component (i.e. DE = 0 in Eq. (5)). A ΔOFV signiﬁcance threshold was calculated to be 16.00 for each single me- tabolite, taking into account the family wise error rate (FWER) using Bonferroni correction. The results were compared to a partial least squares discriminant analysis (PLS-DA) on the data pooled per dose group, using the R-package mixOmics (Cao et al., 2016) after log- transformation and autoscaling of the data (excluding t = 0). A Vari- able Importance in Projection (VIP) on the ﬁrst principal component was calculated for each metabolite. Metabolites with a VIP score > 1 were reported as contributing signi ﬁ cantly to a dose response relation for remoxipride and compared to those selected from the PKPD clus- tering approach. The methods were compared by a weighted Cohen's kappa-analysis.
According to the PLS-DA scores plots, very good clustering was observed for the monocultivar apple juices used in the present study, with juices extracted from Jazz apples showing the largest distance from Granny Smith, Golden Delicious and Pink Lady. As is illustrated in the classiﬁcation matrix for the calibration and validation (testing set) datasets (Table 2), juices produced from Golden Delicious, Jazz, Granny Smith, and Pink Lady apples were 100% correctly classiﬁed whilst in the case of the Braeburn ex- tracted juices only one sample was misclassiﬁed. In both cases the total classiﬁcation percentage was excellent (99.3% and 100% for internal and external validation) which indicates the robust- ness of the PLS-DA predictive models. Moreover, with an RMSE va- lue ranging from 0.10 to 0.23 representing a total error of less than 5%, the predictive power of the herein constructed models is very good. A similar level of performance has previously been seen for geographical characterisation models using a PLS-DA approach constructed with the spectral ﬁngerprint of other DIMS techniques (PTR-MS), applications include agro-industrial products with pro- tected designation of origin such as olive oil, dry cured hams and trufﬂe (Aprea, Biasioli, Carlin, Endrizzi, & Gasperi, 2009; Araghi- pour et al., 2008; Del Pulgar et al., 2011).
Methods: We applied multivariate analysis of PCA, PLS-DA and Cox Regression for clinicopathological features and survival time to explore the significance of miRNA 106b~25 expression in plasma and cancer tissues for gastric cancer. Results: The expression of miRNA 106b, miRNA 93 and miRNA 25 in plasma were positively correlated with their expression in tumor tissues. Via PCA analysis, it was found that miRNA 106b~25 expression in plasma and tumor, T, N and TNM stage were correlated with each other. Via PLS-DA analysis, we identified that T, N and TNM stage were important factors for miRNA 106b~25 expression both in plasma and tumor (all VIP value > 1.2). According to loading weights of variables for the first and second components, it was found that the importance of the miRNA 106b~25s expression carried with the progressed stage of gastric cancer. In the survival analysis, COX regression showed that T stage, plasma miRNA 106b and tumor miRNA 93 were significant risk factors for overall survival [HR: 0.400 (0.205– 0.780); P = 0.007; HR: 0.371 (0.142–0.969), P = 0.043; 0.295 (0.134–0.650), P = 0.002].
plot generated from mass spectrometry spectra of urine samples from dams treated with control (+), combo (x) and caprine milk oligosaccharides (∆) diet. Samples were run in two different columns HILIC and C18 and in both positive and negative ionisation modes separately. (A) HILIC +ve; (B) HILIC -ve; (C) C18 +ve; (D) C18 –ve. ......................... 193 Figure 5.13 The partial least square analysis discriminant analysis (PLS-DA) cross-validated score plot generated from mass spectrometry spectra of urine samples from pups 30 days after weaning from dams treated with control (+), combo (x) and caprine milk oligosaccharides (∆) diet. Samples were run in two different columns HILIC and C18 and in both positive and negative ionisation modes separately. (A) HILIC +ve; (B) HILIC -ve; (C) C18 +ve; (D) C18 –ve. .................................................................................. 196 Figure 5.14 The partial least square analysis discriminant analysis (PLS-DA) cross-validated score plot generated from mass spectrometry spectra of plasma samples from dams treated with control (+), combo (x) and caprine milk oligosaccharides (∆) diet. Samples were run in two different columns HILIC and C18 and in both positive and negative ionisation modes separately. (A) HILIC +ve; (B) HILIC -ve; (C) C18 +ve; (D) C18 –ve. .................................................................................................................................. 200 Figure 5.15 The partial least square analysis discriminant analysis (PLS-DA) cross-validated
The spectral data were exported by Micromass Marker- Lynx ™ applications manager version 4.1 software (Waters Corporation, Milford, MA, USA). The raw data of each sample was normalized to total area to correct for the MS response shift from the first injection to the last injection due to the long duration, overnight or longer, of an LC-MS analysis. The sum of the ion peak area within each sample was set at 10,000. After proces- sing, partial least squares discriminant analysis (PLS-DA) was used for analysis of metabolite profiles, which was performed by the SIMCA-P software version 11 (Ume- trics AB, Umeå, Sweden). In cell culture, each assay was performed in triplicate and repeated three times. The data were expressed as mean ± standard deviation (SD). The significance was tested by one-way analyses of var- iance (ANOVA) of the SPSS 13.0 for Windows (SPSS Inc., Chicago, IL, USA), followed by the Duncan post hoc test. P values less than 0.05 were considered significant.
Abstract: Conventional external beam radiotherapy has been widely used in various clinical malignant and pain management applications. In this study, we developed a serum metabolomic method based on gas chromatogra- phy-mass spectrometry (GC-MS) to evaluate the effect of conventional external beam radiation on rats. Thirty rats were randomly divided to radiation group (600 lx, 800 lx) and control group. Radiation group were under radiation (600 lx, 800 lx) for 1 h. Blood samples were collected from the rats from the control group and radiation group at first, second and third days, respectively. Partial least squares-discriminate analysis (PLS-DA) revealed that radia- tion induced metabolic perturbations. Compared to the control group, the level of propanoic acid of the 600 lx ra- diation group increased; the level of d-Glucose of the 800 lx radiation group decreased at the first day. Compared to the control group, the level of propanoic acid and ethanedioic acid of the 600 lx radiation group increased at the second day. Compared to the control group, the level of propanoic acid of the 600 lx radiation group increased; the level of d-Glucose of the 800 lx radiation group decreased at the first day. The results indicate that metabolomic methods based on GC-MS may be useful to elucidate effect of radiation on rat through the exploration of biomarkers (propanoic acid, d-Glucose, ethanedioic acid).
discriminant models using wood block or tablet samples, these showed the correct prediction rate of 100 %. Using the tablet samples reduced the spectral differences of each species; however, it provided more specific species infor- mation than wood block samples for species classification in the score plots of PLS-DA. In developed models based on wood block and tablet samples, the components of cellulose and water were the main factors affecting species identification. The absorption band at 5464 cm -1 assigned to the semi- or crystalline region in cellulose is considered the most critical component for the identification of P. densiflora and P. sylvestris. We confirmed that a chemo- metric approach combined with NIR spectroscopy and multivariate analysis can be applied for the identification of similar Pinus species.
The vast number of protein spots detected in the 2D- DIGE analysis is evaluated statistically for their signifi- cant higher abundance in either muscle. Different statis- tical approaches can be used on this type of large proteomic data sets, where as the most commonly used statistical method is one-way analysis of variance (ANOVA) (p < 0.05) on individual protein spots. To avoid false positives (type-1 error) due to multiple test- ing a correction for number of conducted tests should be made. Considered as best practice, false discovery rate (FDR) correlation  is applied when approaching proteomic data . In addition, the use of multivariate statistical analysis has the advantage of extracting addi- tional information from the vast data set . In the present study we applied one-way ANOVA statistics with and without FDR correlation and multivariate modelling by partial least square projection to latent structures with discriminant analysis (PLS-DA).
When comparing both loading scatter plots (part B of Figure 3 and Figure 4), the 4 h-extraction method had more complex dots than the 2 h-extraction method. This occurred because in the 4 h-extraction method, the hy- drolysis was longer so that the bioactive compounds could break off from its glucoside and then detected by the instrument. The unknown compounds were labeled as UP no 1 - 9. However, only compounds that have signifi- cant area were included in the determination of marker compounds with PLS-DA method, namely UP4, UP6, and UP9. Because the UP9 was present in all samples, this compound was excluded from marker compounds.A summary of marker compounds of each clone and method were shown in Table 4, while other retention times in loading plots apart from the mentioned compounds were considered as noise.
filing approach. In this study, we found that serum BUN and Scr levels of rat significantly increased, and the renal tissues were damage after a single intraperitoneal injection of cis- platin [20, 21]. Moreover, PNS significantly re- duced the serum BUN and Scr levels, and at- tenuated renal histopathological damages [20, 21]. These results suggest that the model of cisplatin-induced renal damage success and Figure 3. Scores plots (left panel) and cross validation by permutation test (right panel) of PLS-DA derived from the
limit. Hotelling's T-squared statistic is a generalization of Student’s t statistics that is used in multivariate hypoth- esis testing. Two samples were discarded from the data- set due to technical failure during measurement. PLS- DA, a supervised technique, was used to discriminate between non-infected patients versus patients infected with P. aeruginosa or chronically colonized patients ver- sus noncolonized patients. In order to test the perform- ance of the models, a segmented (4 x 7) cross-validation was applied. The quality of the model was evaluated by using the R 2 -value between measured and predicted. The Variable Identification (VID) coefficients were cal- culated to identify possible biomarkers. The VID coeffi- cient was calculated as the correlation coefficient between each original X-variable and the Y-variable as predicted by the PLS-DA model . The VID is there- fore important to understand what the potential rele- vance of each aroma compound is with respect to the predictor variable. PCA and PLS-DA analyses were per- formed using Unscrambler vs 9.8 (CAMO Technologies Inc., Woodbridge, USA).
and the authors ascribed the cytoprotective effect to scavenging dependent antioxidative mechanisms . In the present study on murine myotubes, we revealed an additional antioxidant effect of creatine, i.e. its capacity to induce up-regulation of one of the cellular antioxidative systems the thiol redox system, which consists of the glu- tathione and thioredoxin pathways . Two thioredoxin reductases situated in the mitochondria and cytoplasm, respectively, were increased in creatine treated cells (Table 1); peroxiredoxin-4, a type 2 peroxiredoxin, and thioredoxin dependent peroxide reductase. These sys- tems catalyse thiol-disulfide exchange reactions and Table 1: Relative protein spot volumes of spots identified by PLS-DA