Additional file 4: of CellMapper: rapid and accurate inference of gene expression in difficult-to-isolate cell types
Year: 2016
Type: article
Abstract: Algorithm development, part 2: Parameter optimization for the CellMapper SVD filter, using test searches to find tissue-enriched genes as defined in the TiGER database [55]. a Evaluation of the free parameter, alpha. The SVD filter incorporates a free parameter, alpha, which allows the strength of the filter to be tuned, ranging in value from 1 (weak filter) to 0 (strong filter). Alpha values between 1 and 0.3 led to an increase in AUPR for 25 out of 30 tissues. An intermediate value of 0.5 was chosen for the final algorithm and this parameter was fixed prior to all analyses presented in the main text. b Evaluation of the query-driven weight term (QDW). The SVD filter also includes a term, abbreviated QDW, that decreases the weight of components in which the query genes are not well separated from the rest of the genome. The QDW term leads to an increase in performance beyond what is seen using the alpha scaling factor alone. ***, p < 10−4; Wilcoxon singed rank test. In both subfigures, AUPR was plotted relative to alpha = 1 and no query-driven weight term, which is approximately equivalent to Pearson’s correlation (it is equal to Pearson’s correlation with the low variance principle components filtered, see “Methods”). (PDF 20 kb)
Source: Figshare
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7138109229