This repository summarized the source R Codes of statistical analysis for the article gutjnl-2019-318860, it also contains the detail description of data manipulating process so that others can replicate the results or use the d-signature in clinical practice or validation studies.
System requirements: R >= 3.4.2. Linux system is officially supported, but Mac/Windows should work, too.
To download the new version of the R packages, type the following at the R command line:
install.packages("glmnet")
install.packages("caret")
install.packages("pROC")
install.packages("varSelRF")
install.packages("e1071")
install.packages("Rtsne")
After installing, you have to load to use the package, with: library("glmnet").
For more detailed usage instructions, see the manuscript : gutjnl-2019-318860
| Main processes | Detailed description |
|---|---|
| The specimen collection and exLR-seq bioinformatics analyses. | Five hundred and one participants, PDAC (n=284), CP (n=100), and healthy controls (n = 117) were enrolled. The RNA-seq reads filter perform by FastQC software, the reads mapping evaluated by RnaSeqMap/Hisat2 software. |
| Expression profile normalization, dimensionality reduction and visualization. | The quantification and normalization (TPM) of exLR-seq profiles were manipulated by SanRPKM/FeatureCount software. The dimensional reduction and visualization of variates was conducted by t-SNE algorithm through R package "Rtsne". |
| Different expression analysis and pathway enrichment analysis | The different expression analysis for total samples were performed by Mann-Whitney U test (FDR<0.05, fold change>2). The DEGs were used to perform enrichment annotation of KEGG pathway. |
| Grouping of total samples, potential diagnostic power estimation | Training cohort (Healthy=51; CP=16; PDAC=121), internal validation cohort (Healthy=23; CP=44; PDAC=68), external validation cohort (Healthy=43; CP=40; PDAC=95). The PASS software was used to estimate the potential diagnostic power based on studied samples. |
| Training cohort different expression analysis and feature selection | The different expressed exLRs were selected by Mann-Whitney U test integrated pancreatic tissue and plasma samples (FDR<0.05, fold change>1.4), and LASSO and Random forest algorithm performed by R package "glmnet" and "varSelRF" were used to process feature selection. |
| Diagnostic model construction, parameter tuning and d-signature developing | The eight candidate markers were used to establish classifier by SVM algorithm through "e1071" R package. Five-fold internal cross-validation method was implemented by 'caret' package for SVM parameters tuning. The d-signature was quantitated by predict strength of SVM model output. |
| D-signature validated in internal and external validation cohort. | The training diagnostic model was used to estimate d-signature tested across two validation cohorts and the diagnostic efficacy (Sn, Sp, AUC and accuracy) were estimated by SPSS and R package “pROC”. |