Author Correction: The evolutionary history of lethal metastatic prostate cancer.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Clonal hematopoiesis and risk of prostate cancer in large samples of European ancestry men.

Little is known regarding the potential relationship between clonal hematopoiesis (CH) of indeterminate potential (CHIP), which is the expansion of hematopoietic stem cells with somatic mutations, and risk of prostate cancer, the fifth leading cause of cancer death of men worldwide. We evaluated the association of age-related CHIP with overall and aggressive prostate cancer risk in two large whole-exome sequencing studies of 75 047 European ancestry men, including 7663 prostate cancer cases, 2770 of which had aggressive disease, and 3266 men carrying CHIP variants. We found that CHIP, defined by over 50 CHIP genes individually and in aggregate, was not significantly associated with overall (aggregate HR = 0.93, 95% CI = 0.76-1.13, P = 0.46) or aggressive (aggregate OR = 1.14, 95% CI = 0.92-1.41, P = 0.22) prostate cancer risk. CHIP was weakly associated with genetic risk of overall prostate cancer, measured using a polygenic risk score (OR = 1.05 per unit increase, 95% CI = 1.01-1.10, P = 0.01). CHIP was not significantly associated with carrying pathogenic/likely pathogenic/deleterious variants in DNA repair genes, which have previously been found to be associated with aggressive prostate cancer. While findings from this study suggest that CHIP is likely not a risk factor for prostate cancer, it will be important to investigate other types of CH in association with prostate cancer risk.

Evaluating Germline Testing Panels in Southern African Males With Advanced Prostate Cancer.

BACKGROUND: Germline testing for prostate cancer is on the increase, with clinical implications for risk assessment, treatment, and management. Regardless of family history, NCCN recommends germline testing for patients with metastatic, regional, very-high-risk localized, and high-risk localized prostate cancer. Although African ancestry is a significant risk factor for aggressive prostate cancer, due to a lack of available data no testing criteria have been established for ethnic minorities. PATIENTS AND METHODS: Through deep sequencing, we interrogated the 20 most common germline testing panel genes in 113 Black South African males presenting with largely advanced prostate cancer. Bioinformatic tools were then used to identify the pathogenicity of the variants. RESULTS: After we identified 39 predicted deleterious variants (16 genes), further computational annotation classified 17 variants as potentially oncogenic (12 genes; 17.7% of patients). Rare pathogenic variants included CHEK2 Arg95Ter, BRCA2 Trp31Arg, ATM Arg3047Ter (2 patients), and TP53 Arg282Trp. Notable oncogenic variants of unknown pathogenicity included novel BRCA2 Leu3038Ile in a patient with early-onset disease, whereas patients with FANCA Arg504Cys and RAD51C Arg260Gln reported a family history of prostate cancer. Overall, rare pathogenic and early-onset or familial-associated oncogenic variants were identified in 6.9% (5/72) and 9.2% (8/87) of patients presenting with a Gleason score ≥8 or ≥4 + 3 prostate cancer, respectively. CONCLUSIONS: In this first-of-its-kind study of southern African males, we provide support of African inclusion for advanced, early-onset, and familial prostate cancer genetic testing, indicating clinical value for 30% of current gene panels. Recognizing current panel limitations highlights an urgent need to establish testing guidelines for men of African ancestry. We provide a rationale for considering lowering the pathologic diagnostic inclusion criteria and call for further genome-wide interrogation to ensure the best possible African-relevant prostate cancer gene panel.

Prostate cancer risk in men of differing genetic ancestry and approaches to disease screening and management in these groups.

Prostate cancer is the second most common solid tumour in men worldwide and it is also the most common cancer affecting men of African descent. Prostate cancer incidence and mortality vary across regions and populations. Some of this is explained by a large heritable component of this disease. It has been established that men of African and African Caribbean ethnicity are predisposed to prostate cancer (PrCa) that can have an earlier onset and a more aggressive course, thereby leading to poorer outcomes for patients in this group. Literature searches were carried out using the PubMed, EMBASE and Cochrane Library databases to identify studies associated with PrCa risk and its association with ancestry, screening and management of PrCa. In order to be included, studies were required to be published in English in full-text form. An attractive approach is to identify high-risk groups and develop a targeted screening programme for them as the benefits of population-wide screening in PrCa using prostate-specific antigen (PSA) testing in general population screening have shown evidence of benefit; however, the harms are considered to weigh heavier because screening using PSA testing can lead to over-diagnosis and over-treatment. The aim of targeted screening of higher-risk groups identified by genetic risk stratification is to reduce over-diagnosis and treat those who are most likely to benefit.

The BARCODE1 Pilot: a feasibility study of using germline single nucleotide polymorphisms to target prostate cancer screening.

OBJECTIVES: To assess the feasibility and uptake of a community-based prostate cancer (PCa) screening programme selecting men according to their genetic risk of PCa. To assess the uptake of PCa screening investigations by men invited for screening. The uptake of the pilot study would guide the opening of the larger BARCODE1 study recruiting 5000 men. SUBJECTS AND METHODS: Healthy males aged 55-69 years were invited to participate via their general practitioners (GPs). Saliva samples were collected via mailed collection kits. After DNA extraction, genotyping was conducted using a study specific assay. Genetic risk was based on genotyping 130 germline PCa risk single nucleotide polymorphisms (SNPs). A polygenic risk score (PRS) was calculated for each participant using the sum of weighted alleles for 130 SNPs. Study participants with a PRS lying above the 90th centile value were invited for PCa screening by prostate magnetic resonance imaging (MRI) and biopsy. RESULTS: Invitation letters were sent to 1434 men. The overall study uptake was 26% (375/1436) and 87% of responders were eligible for study entry. DNA genotyping data were available for 297 men and 25 were invited for screening. After exclusions due to medical comorbidity/invitations declined, 18 of 25 men (72%) underwent MRI and biopsy of the prostate. There were seven diagnoses of PCa (38.9%). All cancers were low-risk and were managed with active surveillance. CONCLUSION: The BARCODE1 Pilot has shown this community study in the UK to be feasible, with an overall uptake of 26%. The main BARCODE1 study is now open and will recruit 5000 men. The results of BARCODE1 will be important in defining the role of genetic profiling in targeted PCa population screening. Patient Summary What is the paper about? Very few prostate cancer screening programmes currently exist anywhere in the world. Our pilot study investigated if men in the UK would find it acceptable to have a genetic test based on a saliva sample to examine their risk of prostate cancer development. This test would guide whether men are offered prostate cancer screening tests. What does it mean for patients? We found that the study design was acceptable: 26% of men invited to take part agreed to have the test. The majority of men who were found to have an increased genetic risk of prostate cancer underwent further tests offered (prostate MRI scan and biopsy). We have now expanded the study to enrol 5000 men. The BARCODE1 study will be important in examining whether this approach could be used for large-scale population prostate cancer screening.

African-specific improvement of a polygenic hazard score for age at diagnosis of prostate cancer.

Polygenic hazard score (PHS) models are associated with age at diagnosis of prostate cancer. Our model developed in Europeans (PHS46) showed reduced performance in men with African genetic ancestry. We used a cross-validated search to identify single nucleotide polymorphisms (SNPs) that might improve performance in this population. Anonymized genotypic data were obtained from the PRACTICAL consortium for 6253 men with African genetic ancestry. Ten iterations of a 10-fold cross-validation search were conducted to select SNPs that would be included in the final PHS46+African model. The coefficients of PHS46+African were estimated in a Cox proportional hazards framework using age at diagnosis as the dependent variable and PHS46, and selected SNPs as predictors. The performance of PHS46 and PHS46+African was compared using the same cross-validated approach. Three SNPs (rs76229939, rs74421890 and rs5013678) were selected for inclusion in PHS46+African. All three SNPs are located on chromosome 8q24. PHS46+African showed substantial improvements in all performance metrics measured, including a 75% increase in the relative hazard of those in the upper 20% compared to the bottom 20% (2.47-4.34) and a 20% reduction in the relative hazard of those in the bottom 20% compared to the middle 40% (0.65-0.53). In conclusion, we identified three SNPs that substantially improved the association of PHS46 with age at diagnosis of prostate cancer in men with African genetic ancestry to levels comparable to Europeans.

Marital status and prostate cancer incidence: a pooled analysis of 12 case-control studies from the PRACTICAL consortium.

While being in a committed relationship is associated with a better prostate cancer prognosis, little is known about how marital status relates to its incidence. Social support provided by marriage/relationship could promote a healthy lifestyle and an increased healthcare seeking behavior. We investigated the association between marital status and prostate cancer risk using data from the PRACTICAL Consortium. Pooled analyses were conducted combining 12 case-control studies based on histologically-confirmed incident prostate cancers and controls with information on marital status prior to diagnosis/interview. Marital status was categorized as married/partner, separated/divorced, single, or widowed. Tumours with Gleason scores ≥ 8 defined high-grade cancers, and low-grade otherwise. NCI-SEER's summary stages (local, regional, distant) indicated the extent of the cancer. Logistic regression was used to derive odds ratios (ORs) and 95% confidence intervals (CI) for the association between marital status and prostate cancer risk, adjusting for potential confounders. Overall, 14,760 cases and 12,019 controls contributed to analyses. Compared to men who were married/with a partner, widowed men had an OR of 1.19 (95% CI 1.03-1.35) of prostate cancer, with little difference between low- and high-grade tumours. Risk estimates among widowers were 1.14 (95% CI 0.97-1.34) for local, 1.53 (95% CI 1.22-1.92) for regional, and 1.56 (95% CI 1.05-2.32) for distant stage tumours. Single men had elevated risks of high-grade cancers. Our findings highlight elevated risks of incident prostate cancer among widowers, more often characterized by tumours that had spread beyond the prostate at the time of diagnosis. Social support interventions and closer medical follow-up in this sub-population are warranted.

The evolutionary history of lethal metastatic prostate cancer.

Cancers emerge from an ongoing Darwinian evolutionary process, often leading to multiple competing subclones within a single primary tumour. This evolutionary process culminates in the formation of metastases, which is the cause of 90% of cancer-related deaths. However, despite its clinical importance, little is known about the principles governing the dissemination of cancer cells to distant organs. Although the hypothesis that each metastasis originates from a single tumour cell is generally supported, recent studies using mouse models of cancer demonstrated the existence of polyclonal seeding from and interclonal cooperation between multiple subclones. Here we sought definitive evidence for the existence of polyclonal seeding in human malignancy and to establish the clonal relationship among different metastases in the context of androgen-deprived metastatic prostate cancer. Using whole-genome sequencing, we characterized multiple metastases arising from prostate tumours in ten patients. Integrated analyses of subclonal architecture revealed the patterns of metastatic spread in unprecedented detail. Metastasis-to-metastasis spread was found to be common, either through de novo monoclonal seeding of daughter metastases or, in five cases, through the transfer of multiple tumour clones between metastatic sites. Lesions affecting tumour suppressor genes usually occur as single events, whereas mutations in genes involved in androgen receptor signalling commonly involve multiple, convergent events in different metastases. Our results elucidate in detail the complex patterns of metastatic spread and further our understanding of the development of resistance to androgen-deprivation therapy in prostate cancer.

Genome-wide association study implicates immune dysfunction in the development of Hodgkin lymphoma.

To further our understanding of inherited susceptibility to Hodgkin lymphoma (HL), we performed a meta-analysis of 7 genome-wide association studies totaling 5325 HL cases and 22 423 control patients. We identify 5 new HL risk loci at 6p21.31 (rs649775; P = 2.11 × 10-10), 6q23.3 (rs1002658; P = 2.97 × 10-8), 11q23.1 (rs7111520; P = 1.44 × 10-11), 16p11.2 (rs6565176; P = 4.00 × 10-8), and 20q13.12 (rs2425752; P = 2.01 × 10-8). Integration of gene expression, histone modification, and in situ promoter capture Hi-C data at the 5 new and 13 known risk loci implicates dysfunction of the germinal center reaction, disrupted T-cell differentiation and function, and constitutive NF-κB activation as mechanisms of predisposition. These data provide further insights into the genetic susceptibility and biology of HL.

Appraising the relevance of DNA copy number loss and gain in prostate cancer using whole genome DNA sequence data.

A variety of models have been proposed to explain regions of recurrent somatic copy number alteration (SCNA) in human cancer. Our study employs Whole Genome DNA Sequence (WGS) data from tumor samples (n = 103) to comprehensively assess the role of the Knudson two hit genetic model in SCNA generation in prostate cancer. 64 recurrent regions of loss and gain were detected, of which 28 were novel, including regions of loss with more than 15% frequency at Chr4p15.2-p15.1 (15.53%), Chr6q27 (16.50%) and Chr18q12.3 (17.48%). Comprehensive mutation screens of genes, lincRNA encoding sequences, control regions and conserved domains within SCNAs demonstrated that a two-hit genetic model was supported in only a minor proportion of recurrent SCNA losses examined (15/40). We found that recurrent breakpoints and regions of inversion often occur within Knudson model SCNAs, leading to the identification of ZNF292 as a target gene for the deletion at 6q14.3-q15 and NKX3.1 as a two-hit target at 8p21.3-p21.2. The importance of alterations of lincRNA sequences was illustrated by the identification of a novel mutational hotspot at the KCCAT42, FENDRR, CAT1886 and STCAT2 loci at the 16q23.1-q24.3 loss. Our data confirm that the burden of SCNAs is predictive of biochemical recurrence, define nine individual regions that are associated with relapse, and highlight the possible importance of ion channel and G-protein coupled-receptor (GPCR) pathways in cancer development. We concluded that a two-hit genetic model accounts for about one third of SCNA indicating that mechanisms, such haploinsufficiency and epigenetic inactivation, account for the remaining SCNA losses.

Polygenic risk-tailored screening for prostate cancer: A benefit-harm and cost-effectiveness modelling study.

BACKGROUND: The United States Preventive Services Task Force supports individualised decision-making for prostate-specific antigen (PSA)-based screening in men aged 55-69. Knowing how the potential benefits and harms of screening vary by an individual's risk of developing prostate cancer could inform decision-making about screening at both an individual and population level. This modelling study examined the benefit-harm tradeoffs and the cost-effectiveness of a risk-tailored screening programme compared to age-based and no screening. METHODS AND FINDINGS: A life-table model, projecting age-specific prostate cancer incidence and mortality, was developed of a hypothetical cohort of 4.48 million men in England aged 55 to 69 years with follow-up to age 90. Risk thresholds were based on age and polygenic profile. We compared no screening, age-based screening (quadrennial PSA testing from 55 to 69), and risk-tailored screening (men aged 55 to 69 years with a 10-year absolute risk greater than a threshold receive quadrennial PSA testing from the age they reach the risk threshold). The analysis was undertaken from the health service perspective, including direct costs borne by the health system for risk assessment, screening, diagnosis, and treatment. We used probabilistic sensitivity analyses to account for parameter uncertainty and discounted future costs and benefits at 3.5% per year. Our analysis should be considered cautiously in light of limitations related to our model's cohort-based structure and the uncertainty of input parameters in mathematical models. Compared to no screening over 35 years follow-up, age-based screening prevented the most deaths from prostate cancer (39,272, 95% uncertainty interval [UI]: 16,792-59,685) at the expense of 94,831 (95% UI: 84,827-105,630) overdiagnosed cancers. Age-based screening was the least cost-effective strategy studied. The greatest number of quality-adjusted life-years (QALYs) was generated by risk-based screening at a 10-year absolute risk threshold of 4%. At this threshold, risk-based screening led to one-third fewer overdiagnosed cancers (64,384, 95% UI: 57,382-72,050) but averted 6.3% fewer (9,695, 95% UI: 2,853-15,851) deaths from prostate cancer by comparison with age-based screening. Relative to no screening, risk-based screening at a 4% 10-year absolute risk threshold was cost-effective in 48.4% and 57.4% of the simulations at willingness-to-pay thresholds of GBP£20,000 (US$26,000) and £30,000 ($39,386) per QALY, respectively. The cost-effectiveness of risk-tailored screening improved as the threshold rose. CONCLUSIONS: Based on the results of this modelling study, offering screening to men at higher risk could potentially reduce overdiagnosis and improve the benefit-harm tradeoff and the cost-effectiveness of a prostate cancer screening program. The optimal threshold will depend on societal judgements of the appropriate balance of benefits-harms and cost-effectiveness.

Correction: Rare germline variants in DNA repair genes and the angiogenesis pathway predispose prostate cancer patients to develop metastatic disease.

This article was originally published under the standard License to Publish, but has now been made available under a CC BY 4.0 license. The PDF and HTML versions of the paper have been modified accordingly.

REVEL: An Ensemble Method for Predicting the Pathogenicity of Rare Missense Variants.

The vast majority of coding variants are rare, and assessment of the contribution of rare variants to complex traits is hampered by low statistical power and limited functional data. Improved methods for predicting the pathogenicity of rare coding variants are needed to facilitate the discovery of disease variants from exome sequencing studies. We developed REVEL (rare exome variant ensemble learner), an ensemble method for predicting the pathogenicity of missense variants on the basis of individual tools: MutPred, FATHMM, VEST, PolyPhen, SIFT, PROVEAN, MutationAssessor, MutationTaster, LRT, GERP, SiPhy, phyloP, and phastCons. REVEL was trained with recently discovered pathogenic and rare neutral missense variants, excluding those previously used to train its constituent tools. When applied to two independent test sets, REVEL had the best overall performance (p < 10-12) as compared to any individual tool and seven ensemble methods: MetaSVM, MetaLR, KGGSeq, Condel, CADD, DANN, and Eigen. Importantly, REVEL also had the best performance for distinguishing pathogenic from rare neutral variants with allele frequencies <0.5%. The area under the receiver operating characteristic curve (AUC) for REVEL was 0.046-0.182 higher in an independent test set of 935 recent SwissVar disease variants and 123,935 putatively neutral exome sequencing variants and 0.027-0.143 higher in an independent test set of 1,953 pathogenic and 2,406 benign variants recently reported in ClinVar than the AUCs for other ensemble methods. We provide pre-computed REVEL scores for all possible human missense variants to facilitate the identification of pathogenic variants in the sea of rare variants discovered as sequencing studies expand in scale.

AA9int: SNP interaction pattern search using non-hierarchical additive model set.

Motivation: The use of single nucleotide polymorphism (SNP) interactions to predict complex diseases is getting more attention during the past decade, but related statistical methods are still immature. We previously proposed the SNP Interaction Pattern Identifier (SIPI) approach to evaluate 45 SNP interaction patterns/patterns. SIPI is statistically powerful but suffers from a large computation burden. For large-scale studies, it is necessary to use a powerful and computation-efficient method. The objective of this study is to develop an evidence-based mini-version of SIPI as the screening tool or solitary use and to evaluate the impact of inheritance mode and model structure on detecting SNP-SNP interactions. Results: We tested two candidate approaches: the 'Five-Full' and 'AA9int' method. The Five-Full approach is composed of the five full interaction models considering three inheritance modes (additive, dominant and recessive). The AA9int approach is composed of nine interaction models by considering non-hierarchical model structure and the additive mode. Our simulation results show that AA9int has similar statistical power compared to SIPI and is superior to the Five-Full approach, and the impact of the non-hierarchical model structure is greater than that of the inheritance mode in detecting SNP-SNP interactions. In summary, it is recommended that AA9int is a powerful tool to be used either alone or as the screening stage of a two-stage approach (AA9int+SIPI) for detecting SNP-SNP interactions in large-scale studies. Availability and implementation: The 'AA9int' and 'parAA9int' functions (standard and parallel computing version) are added in the SIPI R package, which is freely available at https://linhuiyi.github.io/LinHY_Software/. Supplementary information: Supplementary data are available at Bioinformatics online.

gsSKAT: Rapid gene set analysis and multiple testing correction for rare-variant association studies using weighted linear kernels.

Next-generation sequencing technologies have afforded unprecedented characterization of low-frequency and rare genetic variation. Due to low power for single-variant testing, aggregative methods are commonly used to combine observed rare variation within a single gene. Causal variation may also aggregate across multiple genes within relevant biomolecular pathways. Kernel-machine regression and adaptive testing methods for aggregative rare-variant association testing have been demonstrated to be powerful approaches for pathway-level analysis, although these methods tend to be computationally intensive at high-variant dimensionality and require access to complete data. An additional analytical issue in scans of large pathway definition sets is multiple testing correction. Gene set definitions may exhibit substantial genic overlap, and the impact of the resultant correlation in test statistics on Type I error rate control for large agnostic gene set scans has not been fully explored. Herein, we first outline a statistical strategy for aggregative rare-variant analysis using component gene-level linear kernel score test summary statistics as well as derive simple estimators of the effective number of tests for family-wise error rate control. We then conduct extensive simulation studies to characterize the behavior of our approach relative to direct application of kernel and adaptive methods under a variety of conditions. We also apply our method to two case-control studies, respectively, evaluating rare variation in hereditary prostate cancer and schizophrenia. Finally, we provide open-source R code for public use to facilitate easy application of our methods to existing rare-variant analysis results.

Prostate cancer risk regions at 8q24 and 17q24 are differentially associated with somatic TMPRSS2:ERG fusion status.

Molecular and epidemiological differences have been described between TMPRSS2:ERG fusion-positive and fusion-negative prostate cancer (PrCa). Assuming two molecularly distinct subtypes, we have examined 27 common PrCa risk variants, previously identified in genome-wide association studies, for subtype specific associations in a total of 1221 TMPRSS2:ERG phenotyped PrCa cases. In meta-analyses of a discovery set of 552 cases with TMPRSS2:ERG data and 7650 unaffected men from five centers we have found support for the hypothesis that several common risk variants are associated with one particular subtype rather than with PrCa in general. Risk variants were analyzed in case-case comparisons (296 TMPRSS2:ERG fusion-positive versus 256 fusion-negative cases) and an independent set of 669 cases with TMPRSS2:ERG data was established to replicate the top five candidates. Significant differences (P < 0.00185) between the two subtypes were observed for rs16901979 (8q24) and rs1859962 (17q24), which were enriched in TMPRSS2:ERG fusion-negative (OR = 0.53, P = 0.0007) and TMPRSS2:ERG fusion-positive PrCa (OR = 1.30, P = 0.0016), respectively. Expression quantitative trait locus analysis was performed to investigate mechanistic links between risk variants, fusion status and target gene mRNA levels. For rs1859962 at 17q24, genotype dependent expression was observed for the candidate target gene SOX9 in TMPRSS2:ERG fusion-positive PrCa, which was not evident in TMPRSS2:ERG negative tumors. The present study established evidence for the first two common PrCa risk variants differentially associated with TMPRSS2:ERG fusion status. TMPRSS2:ERG phenotyping of larger studies is required to determine comprehensive sets of variants with subtype-specific roles in PrCa.

Rare germline variants in DNA repair genes and the angiogenesis pathway predispose prostate cancer patients to develop metastatic disease.

BACKGROUND: Prostate cancer (PrCa) demonstrates a heterogeneous clinical presentation ranging from largely indolent to lethal. We sought to identify a signature of rare inherited variants that distinguishes between these two extreme phenotypes. METHODS: We sequenced germline whole exomes from 139 aggressive (metastatic, age of diagnosis < 60) and 141 non-aggressive (low clinical grade, age of diagnosis ≥60) PrCa cases. We conducted rare variant association analyses at gene and gene set levels using SKAT and Bayesian risk index techniques. GO term enrichment analysis was performed for genes with the highest differential burden of rare disruptive variants. RESULTS: Protein truncating variants (PTVs) in specific DNA repair genes were significantly overrepresented among patients with the aggressive phenotype, with BRCA2, ATM and NBN the most frequently mutated genes. Differential burden of rare variants was identified between metastatic and non-aggressive cases for several genes implicated in angiogenesis, conferring both deleterious and protective effects. CONCLUSIONS: Inherited PTVs in several DNA repair genes distinguish aggressive from non-aggressive PrCa cases. Furthermore, inherited variants in genes with roles in angiogenesis may be potential predictors for risk of metastases. If validated in a larger dataset, these findings have potential for future clinical application.

Height, selected genetic markers and prostate cancer risk: results from the PRACTICAL consortium.

This corrects the article DOI: 10.1038/bjc.2017.231.

Gene and pathway level analyses of germline DNA-repair gene variants and prostate cancer susceptibility using the iCOGS-genotyping array.

This corrects the article DOI: 10.1038/bjc.2016.50.

SNP interaction pattern identifier (SIPI): an intensive search for SNP-SNP interaction patterns.

Motivation: Testing SNP-SNP interactions is considered as a key for overcoming bottlenecks of genetic association studies. However, related statistical methods for testing SNP-SNP interactions are underdeveloped. Results: We propose the SNP Interaction Pattern Identifier (SIPI), which tests 45 biologically meaningful interaction patterns for a binary outcome. SIPI takes non-hierarchical models, inheritance modes and mode coding direction into consideration. The simulation results show that SIPI has higher power than MDR (Multifactor Dimensionality Reduction), AA_Full, Geno_Full (full interaction model with additive or genotypic mode) and SNPassoc in detecting interactions. Applying SIPI to the prostate cancer PRACTICAL consortium data with approximately 21 000 patients, the four SNP pairs in EGFR-EGFR , EGFR-MMP16 and EGFR-CSF1 were found to be associated with prostate cancer aggressiveness with the exact or similar pattern in the discovery and validation sets. A similar match for external validation of SNP-SNP interaction studies is suggested. We demonstrated that SIPI not only searches for more meaningful interaction patterns but can also overcome the unstable nature of interaction patterns. Availability and Implementation: The SIPI software is freely available at http://publichealth.lsuhsc.edu/LinSoftware/ . Contact: hlin1@lsuhsc.edu. Supplementary information: Supplementary data are available at Bioinformatics online.