It is often difficult to obtain sufficient quantity of RNA molecules for gene expression profiling under many practical situations. Amplification from low-input samples may induce artificial signals. We compared the expression measurements of low-input mRNA samples, from 25 pg to 1000 pg mRNA, which were amplified and profiled by Smart-seq, DP-seq and CEL-seq techniques using the Illumina HiSeq 2000 platform, with those of the paired high-input (50 ng) mRNA samples. Even with 1000 pg mRNA input, we found that thousands of genes had at least 2 folds-change of expression levels in the low-input samples compared with the corresponding paired high-input samples. Consequently, a transcriptional signature based on quantitative expression values and determined from high-input RNA samples cannot be applied to low-input samples, and vice versa. In contrast, the within-sample relative expression orderings (REOs) of approximately 90% of all the gene pairs in the high-input samples were maintained in the paired low-input samples with 1000 pg input mRNA molecules. Similar results were observed in the low-input total RNA samples amplified and profiled by the Whole-Genome DASL technique using the Illumina HumanRef-8 v3.0 platform. As a proof of principle, we developed REOs-based signatures from high-input RNA samples for discriminating cancer tissues and showed that they can be robustly applied to low-input RNA samples. REOs-based signatures determined from the high-input RNA samples can be robustly applied to samples profiled with the low-input RNA samples, as low as the 1000 pg and 250 pg input samples but no longer stable in samples with less than 250 pg RNA input to a certain degree.

Due to experimental batch effects, the application of a quantitative transcriptional signature for disease diagnoses commonly requires inter-sample data normalization, which would be hardly applicable under common clinical settings. Many cancers might have qualitative differences with the non-cancer states in the gene expression pattern. Therefore, it is reasonable to explore the power of qualitative diagnostic signatures which are robust against experimental batch effects and other random factors. Firstly, using data of technical replicate samples from the MicroArray Quality Control (MAQC) project, we demonstrated that the low-throughput PCR-based technologies also exist large measurement variations for gene expression even when the samples were measured in the same test site. Then, we demonstrated the critical limitation of low stability for classifiers based on quantitative transcriptional signatures in applications to individual samples through a case study using a support vector machine and a naïve Bayesian classifier to discriminate colorectal cancer tissues from normal tissues. To address this problem, we identified a signature consisting of three gene pairs for discriminating colorectal cancer tissues from non-cancer (normal and inflammatory bowel disease) tissues based on within-sample relative expression orderings (REOs) of these gene pairs. The signature was well verified using 22 independent datasets measured by different microarray and RNA_seq platforms, obviating the need of inter-sample data normalization. Subtle quantitative information of gene expression measurements tends to be unstable under current technical conditions, which will introduce uncertainty to clinical applications of the quantitative transcriptional diagnostic signatures. For diagnosis of disease states with qualitative transcriptional characteristics, the qualitative REO-based signatures could be robustly applied to individual samples measured by different platforms.

Due to the heterogeneity of cancer, identifying differentially methylated (DM) CpG sites between a set of cancer samples and a set of normal samples cannot tell us which patients have methylation aberrations in a particular DM CpG site. We firstly showed that the relative methylation-level orderings (RMOs) of CpG sites within individual normal lung tissues are highly stable but widely disrupted in lung adenocarcinoma tissues. This finding provides the basis of using the RankComp algorithm, previously developed for differential gene expression analysis at the individual level, to identify DM CpG sites in each cancer tissue compared with its own normal state. Briefly, through comparing with the highly stable normal RMOs predetermined in a large collection of samples for normal lung tissues, the algorithm finds those CpG sites whose hyper- or hypo-methylations may lead to the disrupted RMOs of CpG site pairs within a disease sample based on Fisher’s exact test. Evaluated in 59 lung adenocarcinoma tissues with paired adjacent normal tissues, RankComp reached an average precision of 94.26% for individual-level DM CpG sites. Then, after identifying DM CpG sites in each of the 539 lung adenocarcinoma samples from TCGA, we found five and 44 CpG sites hypermethylated and hypomethylated in above 90% of the disease samples, respectively. These findings were validated in 140 publicly available and eight additionally measured paired cancer-normal samples. Gene expression analysis revealed that four of the five genes, HOXA9, TAL1, ATP8A2, ENG and SPARCL1, each harboring one of the five frequently hypermethylated CpG sites within its promoters, were also frequently down-regulated in lung adenocarcinoma. The common DNA methylation aberrations in lung adenocarcinoma tissues may be important for lung adenocarcinoma diagnosis and therapy.

Currently, pathological examination of gastroscopy biopsy specimens is the gold standard for gastric cancer (GC) diagnosis. However, it has a false-negative rate of 10–20% due to inaccurate sampling locations and/or insufficient sampling amount. A signature should be developed to aid the early diagnosis of GC using biopsy specimens even when they are sampled from inaccurate locations. We extracted a robust qualitative transcriptional signature, based on the within-sample relative expression orderings (REOs) of gene pairs, to discriminate both GC tissues and adjacent-normal tissues from non-GC gastritis, intestinal metaplasia and normal gastric tissues. A signature consisting of two gene pairs for GC diagnosis was identified and validated in data of both biopsy specimens and surgical resection specimens pooled from publicly available datasets measured by different laboratories with different platforms. For gastroscopy biopsy specimens, 96.20% of 79 non-GC tissues were correctly identified as non-GC, and 96.84% of 158 GC tissues and six of seven adjacent-normal tissues were correctly identified as GC. For surgical resection specimens, 98.37% of 2560 GC tissues and 97.28% of 221 adjacent-normal tissues were correctly identified as GC. Especially, 97.67% of the 257 GC patients at stage I were exactly diagnosed as GC. We additionally measured 21 GC tissues from seven different GC patients, each with three specimens sampled from three tumor locations with different proportions of the tumor epithelial cell. All these GC tissues were correctly identified as GC, even when the proportion of the tumor epithelial cell was as low as 14%. The qualitative transcriptional signature can distinguish both GC and adjacent-normal tissues from normal, gastritis and intestinal metaplasia tissues of non-GC patients even using inaccurately sampled biopsy specimens, which can be applied robustly at the individual level to aid the early GC diagnosis.

The Connectivity Map (CMAP) database, an important public data source for drug repositioning, archives gene expression profiles from cancer cell lines treated with and without bioactive small molecules. However, there are only one or two technical replicates for each cell line under one treatment condition. For such small-scale data, current fold-changes-based methods lack statistical control in identifying differentially expressed genes (DEGs) in treated cells. Especially, one-to-one comparison may result in too many drug-irrelevant DEGs due to random experimental factors. To tackle this problem, CMAP adopts a pattern-matching strategy to build “connection” between disease signatures and gene expression changes associated with drug treatments. However, many drug-irrelevant genes may blur the “connection” if all the genes are used instead of pre-selected DEGs induced by drug treatments. We applied OneComp, a customized version of RankComp, to identify DEGs in such small-scale cell line datasets. For a cell line, a list of gene pairs with stable relative expression orderings (REOs) were identified in a large collection of control cell samples measured in different experiments and they formed the background stable REOs. When applying OneComp to a small-scale cell line dataset, the background stable REOs were customized by filtering out the gene pairs with reversal REOs in the control samples of the analyzed dataset. In simulated data, the consistency scores of overlapping genes between DEGs identified by OneComp and SAM were all higher than 99%, while the consistency score of the DEGs solely identified by OneComp was 96.85% according to the observed expression difference method. The usefulness of OneComp was exemplified in drug repositioning by identifying phenformin and metformin related genes using small-scale cell line datasets which helped to support them as a potential anti-tumor drug for non-small-cell lung carcinoma, while the pattern-matching strategy adopted by CMAP missed the two connections. The implementation of OneComp is available at https://github.com/pathint/reoa . OneComp performed well in both the simulated and real data. It is useful in drug repositioning studies by helping to find hidden “connections” between drugs and diseases.

Accurate diagnosis of the origin of brain metastases (BMs) is crucial for tailoring an effective therapy to improve patients’ prognosis. BMs of unknown origin account for approximately 2–14% of patients with BMs. Hence, the aim of this study was to identify the original cancer type of BMs based on their DNA methylation profiles. The DNA methylation profiles of glioma (GM), BM, and seven other types of primary cancers were collected. In comparison with GM, the reversal CpG site pairs were identified for each of the seven other types of primary cancers based on the within-sample relative methylation orderings (RMOs) of the CpG sites. Then, using the reversal CpG site pairs, GMs were distinguished from BMs and the seven other types of primary cancers. All 61 of the GM samples were correctly identified as GM. The cancer type was also identified for the non-GM samples. For the seven other types of primary cancers, greater than 93% of samples of each cancer type were correctly identified as their corresponding cancer type, except for breast cancer, which had an 88% accuracy. For 133 BM samples, 132 BM samples were identified as non-GM, and 95% of the 133 BM samples were correctly classified into their corresponding original cancer types. The RMO-based method can accurately identify the origin of BMs, which is important for precision treatment.