Feature selection translates drug response predictors from cell lines to patients.

Targeted therapies and chemotherapies are prevalent in cancer treatment. Identification of predictive markers to stratify cancer patients who will respond to these therapies remains challenging because patient drug response data are limited. As large amounts of drug response data have been generated by cell lines, methods to efficiently translate cell-line-trained predictors to human tumors will be useful in clinical practice. Here, we propose versatile feature selection procedures that can be combined with any classifier. For demonstration, we combined the feature selection procedures with a (linear) logit model and a (non-linear) K-nearest neighbor and trained these on cell lines to result in LogitDA and KNNDA, respectively. We show that LogitDA/KNNDA significantly outperforms existing methods, e.g., a logistic model and a deep learning method trained by thousands of genes, in prediction AUC (0.70-1.00 for seven of the ten drugs tested) and is interpretable. This may be due to the fact that sample sizes are often limited in the area of drug response prediction. We further derive a novel adjustment on the prediction cutoff for LogitDA to yield a prediction accuracy of 0.70-0.93 for seven drugs, including erlotinib and cetuximab, whose pathways relevant to anti-cancer therapies are also uncovered. These results indicate that our methods can efficiently translate cell-line-trained predictors into tumors.

Identification of the common regulators for hepatocellular carcinoma induced by hepatitis B virus X antigen in a mouse model.

Hepatitis B virus X antigen plays an important role in the development of human hepatocellular carcinoma (HCC). The key regulators controlling the temporal downstream gene expression for HCC progression remains unknown. In this study, we took advantage of systems biology approach and analyzed the microarray data of the HBx transgenic mouse as a screening process to identify the differentially expressed genes and applied the software Pathway Studio to identify potential pathways and regulators involved in HCC. Using subnetwork enrichment analysis, we identified five common regulator genes: EDN1, BMP7, BMP4, SPIB and SRC. Upregulation of the common regulators was validated in the other independent HBx transgenic mouse lines. Furthermore, we verified the correlation of their RNA expression levels by using the human HCC samples, and their protein levels by using the human liver disease tissue arrays. EDN1, bone morphogenetic protein (BMP) 4 and BMP7 were upregulated in cirrhosis, BMP4, BMP7 and SRC were further upregulated in hepatocellular or cholangiocellular carcinoma samples. The trend of increasing expression of the common regulators correlates well with the progression of human liver cancer. Overexpression of the common regulators increases the cell viability, promotes migration and invasiveness and enhances the colony formation ability in Hep3B cells. Our approach allows us to identify the critical genes in hepatocarcinogenesis in an HBx-induced mouse model. The validation of the gene expressions in the liver cancer of human patients and their cellular function assays suggests that the identified common regulators may serve as useful molecular targets for the early-stage diagnosis or therapy for HCC.

Modeling and comparing the organization of circular genomes.

MOTIVATION: Most prokaryotic genomes are circular with a single chromosome (called circular genomes), which consist of bacteria and archaea. Orthologous genes (abbreviated as orthologs) are genes directly evolved from an ancestor gene, and can be traced through different species in evolution. Shared orthologs between bacterial genomes have been used to measure their genome evolution. Here, organization of circular genomes is analyzed via distributions of shared orthologs between genomes. However, these distributions are often asymmetric and bimodal; to date, there is no joint distribution to model such data. This motivated us to develop a family of bivariate distributions with generalized von Mises marginals (BGVM) and its statistical inference. RESULTS: A new measure based on circular grade correlation and the fraction of shared orthologs is proposed for association between circular genomes, and a visualization tool developed to depict genome structure similarity. The proposed procedures are applied to eight pairs of prokaryotes separated from domain down to species, and 13 mycoplasma bacteria that are mammalian pathogens belonging to the same genus. We close with remarks on further applications to many features of genomic organization, e.g. shared transcription factor binding sites, between any pair of circular genomes. Thus, the proposed procedures may be applied to identifying conserved chromosome backbones, among others, for genome construction in synthetic biology. AVAILABILITY: All codes of the BGVM procedures and 1000+ prokaryotic genomes are available at http://www.stat.sinica.edu.tw/∼gshieh/bgvm.htm.

Harnessing Synthetic Lethal Interactions for Personalized Medicine.

Two genes are said to have synthetic lethal (SL) interactions if the simultaneous mutations in a cell lead to lethality, but each individual mutation does not. Targeting SL partners of mutated cancer genes can kill cancer cells but leave normal cells intact. The applicability of translating this concept into clinics has been demonstrated by three drugs that have been approved by the FDA to target PARP for tumors bearing mutations in BRCA1/2. This article reviews applications of the SL concept to translational cancer medicine over the past five years. Topics are (1) exploiting the SL concept for drug combinations to circumvent tumor resistance, (2) using synthetic lethality to identify prognostic and predictive biomarkers, (3) applying SL interactions to stratify patients for targeted and immunotherapy, and (4) discussions on challenges and future directions.

A nonlinear correlation measure with applications to gene expression data.

Nonlinear correlation exists in many types of biomedical data. Several types of pairwise gene expression in humans and other organisms show nonlinear correlation across time, e.g., genes involved in human T helper (Th17) cells differentiation, which motivated this study. The proposed procedure, called Kernelized correlation (Kc), first transforms nonlinear data on the plane via a function (kernel, usually nonlinear) to a high-dimensional (Hilbert) space. Next, we plug the transformed data into a classical correlation coefficient, e.g., Pearson's correlation coefficient (r), to yield a nonlinear correlation measure. The algorithm to compute Kc is developed and the R code is provided online. In three simulated nonlinear cases, when noise in data is moderate, Kc with the RBF kernel (Kc-RBF) outperforms Pearson's r and the well-known distance correlation (dCor). However, when noise in data is low, Pearson's r and dCor perform slightly better than (equivalently to) Kc-RBF in Case 1 and 3 (in Case 2); Kendall's tau performs worse than the aforementioned measures in all cases. In Application 1 to discover genes involved in the early Th17 cell differentiation, Kc is shown to detect the nonlinear correlations of four genes with IL17A (a known marker gene), while dCor detects nonlinear correlations of two pairs, and DESeq fails in all these pairs. Next, Kc outperforms Pearson's and dCor, in estimating the nonlinear correlation of negatively correlated gene pairs in yeast cell cycle regulation. In conclusion, Kc is a simple and competent procedure to measure pairwise nonlinear correlations.

H2B ubiquitylation is part of chromatin architecture that marks exon-intron structure in budding yeast.

BACKGROUND: The packaging of DNA into chromatin regulates transcription from initiation through 3' end processing. One aspect of transcription in which chromatin plays a poorly understood role is the co-transcriptional splicing of pre-mRNA. RESULTS: Here we provide evidence that H2B monoubiquitylation (H2BK123ub1) marks introns in Saccharomyces cerevisiae. A genome-wide map of H2BK123ub1 in this organism reveals that this modification is enriched in coding regions and that its levels peak at the transcribed regions of two characteristic subgroups of genes. First, long genes are more likely to have higher levels of H2BK123ub1, correlating with the postulated role of this modification in preventing cryptic transcription initiation in ORFs. Second, genes that are highly transcribed also have high levels of H2BK123ub1, including the ribosomal protein genes, which comprise the majority of intron-containing genes in yeast. H2BK123ub1 is also a feature of introns in the yeast genome, and the disruption of this modification alters the intragenic distribution of H3 trimethylation on lysine 36 (H3K36me3), which functionally correlates with alternative RNA splicing in humans. In addition, the deletion of genes encoding the U2 snRNP subunits, Lea1 or Msl1, in combination with an htb-K123R mutation, leads to synthetic lethality. CONCLUSION: These data suggest that H2BK123ub1 facilitates cross talk between chromatin and pre-mRNA splicing by modulating the distribution of intronic and exonic histone modifications.

WebPARE: web-computing for inferring genetic or transcriptional interactions.

SUMMARY: Inferring genetic or transcriptional interactions, when done successfully, may provide insights into biological processes or biochemical pathways of interest. Unfortunately, most computational algorithms require a certain level of programming expertise. To provide a simple web interface for users to infer interactions from time course gene expression data, we present WebPARE, which is based on the pattern recognition algorithm (PARE). For expression data, in which each type of interaction (e.g. activator target) and the corresponding paired gene expression pattern are significantly associated, PARE uses a non-linear score to classify gene pairs of interest into a few subclasses of various time lags. In each subclass, PARE learns the parameters in the decision score using known interactions from biological experiments or published literature. Subsequently, the trained algorithm predicts interactions of a similar nature. Previously, PARE was shown to infer two sets of interactions in yeast successfully. Moreover, several predicted genetic interactions coincided with existing pathways; this indicates the potential of PARE in predicting partial pathway components. Given a list of gene pairs or genes of interest and expression data, WebPARE invokes PARE and outputs predicted interactions and their networks in directed graphs.

Immunohistochemical Expression of Five Protein Combinations Revealed as Prognostic Markers in Asian Oral Cancer.

Oral squamous cell carcinoma (OSCC) has a high mortality rate (∼50%), and the 5-year overall survival rate is not optimal. Cyto- and histopathological examination of cancer tissues is the main strategy for diagnosis and treatment. In the present study, we aimed to uncover immunohistochemical (IHC) markers for prognosis in Asian OSCC. From the collected 742 synthetic lethal gene pairs (of various cancer types), we first filtered genes relevant to OSCC, performed 29 IHC stains at different cellular portions and combined these IHC stains into 398 distinct pairs. Next, we identified novel IHC prognostic markers in OSCC among Taiwanese population, from the single and paired IHC staining by univariate Cox regression analysis. Increased nuclear expression of RB1 [RB1(N)↑], CDH3(C)↑-STK17A(N)↑ and FLNA(C)↑-KRAS(C)↑were associated with survival, but not independent of tumor stage, where C and N denote cytoplasm and nucleus, respectively. Furthermore, multivariate Cox regression analyses revealed that CSNK1E(C)↓-SHC1(N)↓ (P = 5.9 × 10-5; recommended for clinical use), BRCA1(N)↓-SHC1(N)↓ (P = 0.030), CSNK1E(C)↓-RB1(N)↑ (P = 0.045), [CSNK1E(C)-SHC1(N), FLNA(C)-KRAS(C)] (P = 0.000, rounded to three decimal places) and [BRCA1(N)-SHC1(N), FLNA(C)-KRAS(C)] (P = 0.020) were significant factors of poor prognosis, independent of lymph node metastasis, stage and alcohol consumption. An external dataset from The Cancer Genome Atlas HNSCC cohort confirmed that CDH3↑-STK17A↑ was a significant predictor of poor survival. Our approach identified prognostic markers with components involved in different pathways and revealed IHC marker pairs while neither single IHC was a marker, thus it improved the current state-of-the-art for identification of IHC markers.

Uncovering transcriptional interactions via an adaptive fuzzy logic approach.

BACKGROUND: To date, only a limited number of transcriptional regulatory interactions have been uncovered. In a pilot study integrating sequence data with microarray data, a position weight matrix (PWM) performed poorly in inferring transcriptional interactions (TIs), which represent physical interactions between transcription factors (TF) and upstream sequences of target genes. Inferring a TI means that the promoter sequence of a target is inferred to match the consensus sequence motifs of a potential TF, and their interaction type such as AT or RT is also predicted. Thus, a robust PWM (rPWM) was developed to search for consensus sequence motifs. In addition to rPWM, one feature extracted from ChIP-chip data was incorporated to identify potential TIs under specific conditions. An interaction type classifier was assembled to predict activation/repression of potential TIs using microarray data. This approach, combining an adaptive (learning) fuzzy inference system and an interaction type classifier to predict transcriptional regulatory networks, was named AdaFuzzy. RESULTS: AdaFuzzy was applied to predict TIs using real genomics data from Saccharomyces cerevisiae. Following one of the latest advances in predicting TIs, constrained probabilistic sparse matrix factorization (cPSMF), and using 19 transcription factors (TFs), we compared AdaFuzzy to four well-known approaches using over-representation analysis and gene set enrichment analysis. AdaFuzzy outperformed these four algorithms. Furthermore, AdaFuzzy was shown to perform comparably to 'ChIP-experimental method' in inferring TIs identified by two sets of large scale ChIP-chip data, respectively. AdaFuzzy was also able to classify all predicted TIs into one or more of the four promoter architectures. The results coincided with known promoter architectures in yeast and provided insights into transcriptional regulatory mechanisms. CONCLUSION: AdaFuzzy successfully integrates multiple types of data (sequence, ChIP, and microarray) to predict transcriptional regulatory networks. The validated success in the prediction results implies that AdaFuzzy can be applied to uncover TIs in yeast.

A pattern recognition approach to infer time-lagged genetic interactions.

MOTIVATION: For any time-course microarray data in which the gene interactions and the associated paired patterns are dependent, the proposed pattern recognition (PARE) approach can infer time-lagged genetic interactions, a challenging task due to the small number of time points and large number of genes. PARE utilizes a non-linear score to identify subclasses of gene pairs with different time lags. In each subclass, PARE extracts non-linear characteristics of paired gene-expression curves and learns weights of the decision score applying an optimization algorithm to microarray gene-expression data (MGED) of some known interactions, from biological experiments or published literature. Namely, PARE integrates both MGED and existing knowledge via machine learning, and subsequently predicts the other genetic interactions in the subclass. RESULTS: PARE, a time-lagged correlation approach and the latest advance in graphical Gaussian models were applied to predict 112 (132) pairs of TC/TD (transcriptional regulatory) interactions. Checked against qRT-PCR results (published literature), their true positive rates are 73% (77%), 46% (51%), and 52% (59%), respectively. The false positive rates of predicting TC and TD (AT and RT) interactions in the yeast genome are bounded by 13 and 10% (10 and 14%), respectively. Several predicted TC/TD interactions are shown to coincide with existing pathways involving Sgs1, Srs2 and Mus81. This reinforces the possibility of applying genetic interactions to predict pathways of protein complexes. Moreover, some experimentally testable gene interactions involving DNA repair are predicted. AVAILABILITY: Supplementary data and PARE software are available at http://www.stat.sinica.edu.tw/~gshieh/pare.htm.

Inferring transcriptional compensation interactions in yeast via stepwise structure equation modeling.

BACKGROUND: With the abundant information produced by microarray technology, various approaches have been proposed to infer transcriptional regulatory networks. However, few approaches have studied subtle and indirect interaction such as genetic compensation, the existence of which is widely recognized although its mechanism has yet to be clarified. Furthermore, when inferring gene networks most models include only observed variables whereas latent factors, such as proteins and mRNA degradation that are not measured by microarrays, do participate in networks in reality. RESULTS: Motivated by inferring transcriptional compensation (TC) interactions in yeast, a stepwise structural equation modeling algorithm (SSEM) is developed. In addition to observed variables, SSEM also incorporates hidden variables to capture interactions (or regulations) from latent factors. Simulated gene networks are used to determine with which of six possible model selection criteria (MSC) SSEM works best. SSEM with Bayesian information criterion (BIC) results in the highest true positive rates, the largest percentage of correctly predicted interactions from all existing interactions, and the highest true negative (non-existing interactions) rates. Next, we apply SSEM using real microarray data to infer TC interactions among (1) small groups of genes that are synthetic sick or lethal (SSL) to SGS1, and (2) a group of SSL pairs of 51 yeast genes involved in DNA synthesis and repair that are of interest. For (1), SSEM with BIC is shown to outperform three Bayesian network algorithms and a multivariate autoregressive model, checked against the results of qRT-PCR experiments. The predictions for (2) are shown to coincide with several known pathways of Sgs1 and its partners that are involved in DNA replication, recombination and repair. In addition, experimentally testable interactions of Rad27 are predicted. CONCLUSION: SSEM is a useful tool for inferring genetic networks, and the results reinforce the possibility of predicting pathways of protein complexes via genetic interactions.

Detection of Somatic Mutations in Exome Sequencing of Tumor-only Samples.

Due to lack of normal samples in clinical diagnosis and to reduce costs, detection of small-scale mutations from tumor-only samples is required but remains relatively unexplored. We developed an algorithm (GATKcan) augmenting GATK with two statistics and machine learning to detect mutations in cancer. The averaged performance of GATKcan in ten experiments outperformed GATK in detecting mutations of randomly sampled 231 from 241 TCGA endometrial tumors (EC). In external validations, GATKcan outperformed GATK in TCGA breast cancer (BC), ovarian cancer (OC) and melanoma tumors, in terms of Matthews correlation coefficient (MCC) and precision, where MCC takes both sensitivity and specificity into account. Further, GATKcan reduced high fractions of false positives detected by GATK. In mutation detection of somatic variants, classified commonly by VarScan 2 and MuTect from the called variants in BC, OC and melanoma, ranked by adjusted MCC (adjusted precision) GATKcan was the top 1, followed by MuTect, VarScan 2 and GATK. Importantly, GATKcan enables detection of mutations when alternate alleles exist in normal samples. These results suggest that GATKcan trained by a cancer is able to detect mutations in future patients with the same type of cancer and is likely applicable to other cancers with similar mutations.

Uncovering synthetic lethal interactions for therapeutic targets and predictive markers in lung adenocarcinoma.

Two genes are called synthetic lethal (SL) if their simultaneous mutation leads to cell death, but mutation of either individual does not. Targeting SL partners of mutated cancer genes can selectively kill cancer cells, but leave normal cells intact. We present an integrated approach to uncover SL gene pairs as novel therapeutic targets of lung adenocarcinoma (LADC). Of 24 predicted SL pairs, PARP1-TP53 was validated by RNAi knockdown to have synergistic toxicity in H1975 and invasive CL1-5 LADC cells; additionally FEN1-RAD54B, BRCA1-TP53, BRCA2-TP53 and RB1-TP53 were consistent with the literature. While metastasis remains a bottleneck in cancer treatment and inhibitors of PARP1 have been developed, this result may have therapeutic potential for LADC, in which TP53 is commonly mutated. We also demonstrated that silencing PARP1 enhanced the cell death induced by the platinum-based chemotherapy drug carboplatin in lung cancer cells (CL1-5 and H1975). IHC of RAD54B↑, BRCA1↓-RAD54B↑, FEN1(N)↑-RAD54B↑ and PARP1↑-RAD54B↑ were shown to be prognostic markers for 131 Asian LADC patients, and all markers except BRCA1↓-RAD54B↑ were further confirmed by three independent gene expression data sets (a total of 426 patients) including The Cancer Genome Atlas (TCGA) cohort of LADC. Importantly, we identified POLB-TP53 and POLB as predictive markers for the TCGA cohort (230 subjects), independent of age and stage. Thus, POLB and POLB-TP53 may be used to stratify future non-Asian LADC patients for therapeutic strategies.

The Overexpression of FEN1 and RAD54B May Act as Independent Prognostic Factors of Lung Adenocarcinoma.

Synthetic lethality arises when a combination of mutations in two or more genes leads to cell death. However, the prognostic role of concordant overexpression of synthetic lethality genes in protein level rather than a combination of mutations is not clear. In this study, we explore the prognostic role of combined overexpression of paired genes in lung adenocarcinoma. We used immunohistochemical staining to investigate 24 paired genes in 93 lung adenocarcinoma patients and Kaplan-Meier analysis and Cox proportional hazards models to evaluate their prognostic roles. Among 24 paired genes, only FEN1 (Flap endonuclease 1) and RAD54B (RAD54 homolog B) were overexpressed in lung adenocarcinoma patients with poor prognosis. Patients with expression of both FEN1 and RAD54B were prone to have advanced nodal involvement and significantly poor prognosis (HR = 2.35, P = 0.0230). These results suggest that intensive follow up and targeted therapy might improve clinical outcome for patients who show expression of both FEN1 and RAD54B.

CSNK1E/CTNNB1 are synthetic lethal to TP53 in colorectal cancer and are markers for prognosis.

Two genes are called synthetic lethal (SL) if their simultaneous mutations lead to cell death, but each individual mutation does not. Targeting SL partners of mutated cancer genes can kill cancer cells specifically, but leave normal cells intact. We present an integrated approach to uncovering SL pairs in colorectal cancer (CRC). Screening verified SL pairs using microarray gene expression data of cancerous and normal tissues, we first identified potential functionally relevant (simultaneously differentially expressed) gene pairs. From the top-ranked pairs, ~20 genes were chosen for immunohistochemistry (IHC) staining in 171 CRC patients. To find novel SL pairs, all 169 combined pairs from the individual IHC were synergistically correlated to five clinicopathological features, e.g. overall survival. Of the 11 predicted SL pairs, MSH2-POLB and CSNK1E-MYC were consistent with literature, and we validated the top two pairs, CSNK1E-TP53 and CTNNB1-TP53 using RNAi knockdown and small molecule inhibitors of CSNK1E in isogenic HCT-116 and RKO cells. Furthermore, synthetic lethality of CSNK1E and TP53 was verified in mouse model. Importantly, multivariate analysis revealed that CSNK1E-P53, CTNNB1-P53, MSH2-RB1, and BRCA1-WNT5A were independent prognosis markers from stage, with CSNK1E-P53 applicable to early-stage and the remaining three throughout all stages. Our findings suggest that CSNK1E is a promising target for TP53-mutant CRC patients which constitute ~40% to 50% of patients, while to date safety regarding inhibition of TP53 is controversial. Thus the integrated approach is useful in finding novel SL pairs for cancer therapeutics, and it is readily accessible and applicable to other cancers.

Inferring coregulation of transcription factors and microRNAs in breast cancer.

Both transcription factors (TFs) and microRNAs (miRNAs) regulate gene expression. TFs activate or suppress the initiation of the transcription process and miRNAs regulate mRNAs post-transcriptionally, thus forming a temporally ordered regulatory event. Ectopic expression of key transcriptional regulators and/or miRNAs has been shown to be involved in various tumors. Therefore, uncovering the coregulation of TFs and miRNAs in human cancers may lead to the discovery of novel therapeutics. We introduced a two-stage learning fuzzy method to model TF-miRNA coregulation using both genomic data and verified regulatory relationships. In Stage 1, a learning (adaptive) fuzzy inference system (ANFIS) combines two sequence alignment features of TF and target by learning from verified TF-target pairs into a sequence matching score. Next, a non-learning FIS incorporates a sequence alignment score and a correlation score from paired TF-target gene expression to output a Stage 1 fuzzy score to infer whether a TF-target regulation exists. For significant TF-target pairs, in Stage 2, similar to Stage 1, we first infer whether a miRNA regulates each common target by an ANFIS, which incorporates their sequences and known miRNA-target relationships to output a sequence score. Next, an FIS incorporates the Stage 1 fuzzy score, Stage 2 sequence score and gene expression correlation score of a miRNA-target pair to determine whether TF-miRNA coregulation exists. We collected 54 (8) TF-miRNA-target triples validated in ER-positive (ER-negative) breast cancer cell lines in the same article, and they were used as positives. Negative examples were constructed for Stage 1 (Stage 2) by pairing TFs (miRNAs) with human housekeeping genes found in the literature; both positives and negatives were used to train ANFISs in the training step. This two-stage fizzy algorithm was applied to predict 54 (8) TF-miRNA coregulation triples in ER-positive (ER-negative) human breast cancer cell lines, and resulted in true-positive rates of 0.55 (0.74) and 0.57 (0.75) using 3-fold and n-fold cross validations (CVs), respectively. False-positive rate bound was 0.07 (0.13) for ER-positive (ER-negative) breast cancer using both 3-fold and n-fold CVs. Interestingly, among the 62 coregulatroy triples from ER-positive/negative breast cancer cells, about 72% have TF- and coregulatory miRNA expression simultaneously greater or less than the corresponding medians, while in the remaining 28% of TFs and their coregulatory miRNAs are conversely expressed. The proposed fuzzy algorithm performed well in identification of TF-miRNA coregulation triples in human breast cancer. After being trained by the corresponding verified coregulatory triples and genomic data, this algorithm can be applied to uncover novel coregulation in other cancers in the future.

Inferring Genetic Interactions via a Data-Driven Second Order Model.

Genetic/transcriptional regulatory interactions are shown to predict partial components of signaling pathways, which have been recognized as vital to complex human diseases. Both activator (A) and repressor (R) are known to coregulate their common target gene (T). Xu et al. (2002) proposed to model this coregulation by a fixed second order response surface (called the RS algorithm), in which T is a function of A, R, and AR. Unfortunately, the RS algorithm did not result in a sufficient number of genetic interactions (GIs) when it was applied to a group of 51 yeast genes in a pilot study. Thus, we propose a data-driven second order model (DDSOM), an approximation to the non-linear transcriptional interactions, to infer genetic and transcriptional regulatory interactions. For each triplet of genes of interest (A, R, and T), we regress the expression of T at time t + 1 on the expression of A, R, and AR at time t. Next, these well-fitted regression models (viewed as points in R(3)) are collected, and the center of these points is used to identify triples of genes having the A-R-T relationship or GIs. The DDSOM and RS algorithms are first compared on inferring transcriptional compensation interactions of a group of yeast genes in DNA synthesis and DNA repair using microarray gene expression data; the DDSOM algorithm results in higher modified true positive rate (about 75%) than that of the RS algorithm, checked against quantitative RT-polymerase chain reaction results. These validated GIs are reported, among which some coincide with certain interactions in DNA repair and genome instability pathways in yeast. This suggests that the DDSOM algorithm has potential to predict pathway components. Further, both algorithms are applied to predict transcriptional regulatory interactions of 63 yeast genes. Checked against the known transcriptional regulatory interactions queried from TRANSFAC, the proposed also performs better than the RS algorithm.

Inferring genetic interactions via a nonlinear model and an optimization algorithm.

BACKGROUND: Biochemical pathways are gradually becoming recognized as central to complex human diseases and recently genetic/transcriptional interactions have been shown to be able to predict partial pathways. With the abundant information made available by microarray gene expression data (MGED), nonlinear modeling of these interactions is now feasible. Two of the latest advances in nonlinear modeling used sigmoid models to depict transcriptional interaction of a transcription factor (TF) for a target gene, but do not model cooperative or competitive interactions of several TFs for a target. RESULTS: An S-shape model and an optimization algorithm (GASA) were developed to infer genetic interactions/transcriptional regulation of several genes simultaneously using MGED. GASA consists of a genetic algorithm (GA) and a simulated annealing (SA) algorithm, which is enhanced by a steepest gradient descent algorithm to avoid being trapped in local minimum. Using simulated data with various degrees of noise, we studied how GASA with two model selection criteria and two search spaces performed. Furthermore, GASA was shown to outperform network component analysis, the time series network inference algorithm (TSNI), GA with regular GA (GAGA) and GA with regular SA. Two applications are demonstrated. First, GASA is applied to infer a subnetwork of human T-cell apoptosis. Several of the predicted interactions are supported by the literature. Second, GASA was applied to infer the transcriptional factors of 34 cell cycle regulated targets in S. cerevisiae, and GASA performed better than one of the latest advances in nonlinear modeling, GAGA and TSNI. Moreover, GASA is able to predict multiple transcription factors for certain targets, and these results coincide with experiments confirmed data in YEASTRACT. CONCLUSIONS: GASA is shown to infer both genetic interactions and transcriptional regulatory interactions well. In particular, GASA seems able to characterize the nonlinear mechanism of transcriptional regulatory interactions (TIs) in yeast, and may be applied to infer TIs in other organisms. The predicted genetic interactions of a subnetwork of human T-cell apoptosis coincide with existing partial pathways, suggesting the potential of GASA on inferring biochemical pathways.