Average variable importance in the projection (VIP) scores from the PLSR models analyzing the covariation of molecular signals with cell viability and apoptosis fraction in BRAF(V600E/D) melanoma cell lines - Dataset (ID:20231)
|HMS Dataset ID:||20231|
|Dataset Title:||Average variable importance in the projection (VIP) scores from the PLSR models analyzing the covariation of molecular signals with cell viability and apoptosis fraction in BRAF(V600E/D) melanoma cell lines|
|Publication(s) Using Dataset:||PMID: 25814555|
|Project Summary Page(s):||lincs.hms.harvard.edu/fallahi-sichani-molsystbiol-2015|
|Screening Lab Investigator:||Mohammad Fallahi-Sichani|
|Screening Principal Investigator:||Peter Sorger|
|Assay Description:||This dataset presents an analysis of the variable importance in the projection (VIP) scores for each variable in the partial-least-squares regression (PLSR) models arising from analysis of HMS LINCS datasets #20217 and 20218, which assess covariation of molecular signals (measured by RPPA; dataset #20218) with cellular responses (relative viability and apoptotic fractions; dataset #20217). pMEK and pERK signals were excluded from the analysis.|
1. Cellular responses at the level of relative viability, apoptotic fraction, and protein phosphorylation were assessed as described in datasets #20217 and 20218.|
2. Partial-least-squares regression (PLSR) modeling was used to identify statistically significant covariation between molecular signals (input data; measured by RPPA) and corresponding cellular responses (output data; relative viability and apoptotic fractions) for each cell line, as described in datasets #20229 and 20230.
3. For assessment of relative variable importance in each PLSR model, the information content of each variable (representing a signal measurement at a specific time point) was assessed by its variable importance in the projection (VIP) (Janes et al, 2008;Wold, 1994):
where K is the total number of signaling variables (K = 21×5 = 105), wnk is the weight of the kth variable for the nth PLSR component, N is the total number of PLSR components, and SSn is the sum of squares explained by the nth PLSR component. VIP scores were generated for PLSR models using RPPA signals excluding pMEK and pERK measurements (as changes in these signals are a direct consequence of drug action), following the specifications given below.
4. RPPA slides were scanned and analyzed twice using two different scanners and two different image analysis programs.
5. The two RPPA data sets were used independently to generate data-driven models for each cell line.
6. The signal/time-point measurements that did not show consistent up or down-regulation between the two analyses were removed. Overall, ~75% of the VIP data were consistent between the two analyses. Most (~57%) of the inconsistent VIP scores (i.e. scores that had different signs between the two analyses) were insignificant (|VIP| < 1) in both analyses, and ~37% of them were insignificant in at least one of the analyses. Only 6% of the significant VIP scores from the two analyses were not consistent. Nevertheless, all of the inconsistent data were removed from further analysis (specified as “NaN” in the dataset).
7. Model-derived VIP scores for the remaining data were averaged between the two datasets.
|Assay Protocol Reference:||
Janes KA, Reinhardt HC, Yaffe MB (2008) Cytokine-induced signaling networks prioritize dynamic range over signal strength. Cell 135:343-354.|
Wold S (1994) Exponentially weighted moving principal components analysis and projections to latent structures. Chemometrics Intellig Lab Syst 23:149-161.
|HMS Dataset Type:||Analysis|
|Date Publicly Available:||2015-04-17|
|Most Recent Update:||2015-09-23|