Performance evaluation for rapid detection of pan-cancer microsatellite instability with MANTIS.

Kautto EA, Bonneville R, Miya J, Yu L, Krook MA, Reeser JW, Roychowdhury S. · 2017

Read Performance evaluation for rapid detection of pan-cancer microsatellite instability with MANTIS. by Kautto EA, Bonneville R, Miya J, Yu L, Krook MA, Reeser JW, Roychowdhury S. (2017) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc5352334.txt.

Introduction

Esko A. Kautto and co-authors tested a new tool for a cancer mark in Oncotarget in 2017. Some tumors lose their DNA spell-check repair, which leaves repeat tracts unstable. That mark, called microsatellite instability, guides care and drug choice. Lab tests for it are slow and hands-on. Gene reading could speed it, but no fair race of the reading tools had been run.

Their tool, MANTIS, was set against two past tools on more than four hundred and fifty tumor and normal pairs from six cancer types. All three tools scored well. MANTIS came top, with the best blend of catching true cases and clearing true negatives, and it held that lead on small and large gene panels alike. It also ran with the least use of machine power. The team offers it as a fast, trusty screen for the mark across cancers.

Catalog record

CREATORKautto EA, Bonneville R, Miya J, Yu L, Krook MA, Reeser JW, Roychowdhury S.
YEAR2017
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC5352334

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Performance evaluation for rapid detection of pan-cancer microsatellite instability with MANTIS By Esko A. Kautto, Russell Bonneville, Jharna Miya, Lianbo Yu, Melanie A. Krook, Julie W. Reeser, Sameek Roychowdhury Published in: Oncotarget DOI: 10.18632/oncotarget.13918 ABSTRACT In current clinical practice, microsatellite instability (MSI) and mismatch repair deficiency detection is performed with MSI-PCR and immunohistochemistry. Recent research has produced several computational tools for MSI detection with next-generation sequencing (NGS) data; however a comprehensive analysis of computational methods has not yet been performed. In this study, we introduce a new MSI detection tool, MANTIS, and demonstrate its favorable performance compared to the previously published tools mSINGS and MSISensor. We evaluated 458 normal-tumor sample pairs across six cancer subtypes, testing classification performance on variable numbers of target loci ranging from 10 to 2539. All three computational methods were found to be accurate, with MANTIS exhibiting the highest accuracy with 98.91% of samples from all six diseases classified correctly. MANTIS displayed superior performance among the three tools, having the highest overall sensitivity (MANTIS 97.18%, MSISensor 96.48%, mSINGS 76.06%) and specificity (MANTIS 99.68%, mSINGS 99.68%, MSISensor 98.73%) across six cancer types, even with loci panels of varying size. Additionally, MANTIS also had the lowest resource consumption (<1% of the space and <7% of the memory required by mSINGS) and fastest running times (49.6% and 8.7% of the running times of MSISensor and mSINGS, respectively). This study highlights the potential utility of MANTIS in classifying samples by MSI-status, allowing its incorporation into existing NGS pipelines. INTRODUCTION Microsatellites are short (1-6bp) repeating motifs, widely dispersed throughout the human genome [1]. Microsatellite instability (MSI) is a genetic phenomenon of somatic polymorphisms of microsatellite length, caused by uncorrected “slippage” of DNA fragments during DNA replication in cell division. MSI can arise from defects in the DNA mismatch repair (MMR) system [2]. These defects may be inherited, as with Lynch syndrome/hereditary nonpolyposis colorectal cancer [3], or may be somatically acquired, most commonly due to promoter hypermethylation of the MMR gene MLH1 [4]. Increasing evidence demonstrates that MSI is a recurrent somatic abnormality in several human cancers, found in 13% of colorectal adenocarcinoma and 22-33% of uterine/endometrial carcinoma [5]. Reliable detection of MSI is clinically useful as MSI-positive tumors appear more susceptible to immune-enhancing therapies, as observed in colorectal cancer for the PD-1 inhibitor pembrolizumab [6]. The two currently accepted assays for the detection of MSI are MSI-PCR of five standardized microsatellite loci (Bethesda panel) [7], and immunohistochemistry (IHC) of the MMR proteins MSH2, MSH6, MLH1 and PMS2. Traditionally, tumors can be classified with MSI-PCR as microsatellite stable (MSS, 0/5 loci unstable), MSI-low (MSI-L, 1/5 loci unstable), or MSI-high (MSI-H, ≥ 2/5 loci unstable) [7]. However, both of these methods have inherent limitations. MSI-PCR relies on a small set of loci that were selected based on markers from a single disease type, potentially excluding loci that would be better predictors in other diseases and increasing the odds of incorrect classification [7]. Immunohistochemistry can be used to detect the expression of mismatch repair proteins, but does not directly look at the microsatellite loci [7]. More recently, with the increasing prevalence of next-generation sequencing (NGS) in cancer biology, several computational methods have been developed using either colorectal or endometrial cancer NGS data to determine MSI status [8, 9]. The development, refinement, and validation of NGS-based computational MSI calling methods have several research and clinical applications. NGS allows for the practical assessment of far more microsatellite loci than MSI-PCR. Importantly, computational MSI analyses can be integrated into existing NGS pipelines for other mutation types such as single nucleotide variation or copy number variation, as well as applied to previously generated NGS data for retrospective analyses. As NGS increases in both cost-effectiveness and prevalence, NGS-based methods may permit identification of MSI status without requiring additional clinical testing or patient sample processing. Lastly, NGS data is becoming increasingly available in tumor types that are not routinely tested for MSI, with potential opportunities to identify microsatellite instability in previously uncharacterized cancers. Thus, there is a need to develop tools with high accuracy in multiple cancer types. In this study, we introduce a new tool, MANTIS (Microsatellite Analysis for Normal Tumor InStability), for detecting MSI status from NGS data. We compare MANTIS with two currently available tools, mSINGS [8] and MSISensor [9], and test their performance across six different tumor types. We also determine that the number of loci assessed impacts the accuracy of MSI calling methods, and find an optimal number of loci to analyze with each tool for best tool performance. RESULTS Evaluation of tool accuracy for detecting MSI status Since mSINGS and MSISensor were each developed on data from single disease cohorts (COAD/READ for mSINGS, UCEC for MSISensor), the two cohorts were used for tool performance comparisons. To account for the possibility of suboptimal recommended cutoff thresholds for each of the three tools, we tested a range of thresholds for each tool across a test cohort consisting of both the COAD/READ and UCEC sample pairs (Supplemental Figure S1). The peak performances of each tool were determined, with MANTIS having 97.1% accuracy with the default threshold of 0.4, mSINGS reaching 96% accuracy with a threshold of 0.1, and MSISensor peaking at 95.4% accuracy with the threshold of 3.5%. The results indicate that MANTIS demonstrated superior performance over the other tools, even after accounting for the best-case thresholds of the tools. Having evaluated the tools with the best-case thresholds, their performances were evaluated with the tool's recommended default cutoff thresholds (MANTIS 0.4, mSINGS 0.2, and MSISensor 3.5%) (Supplemental Figure S2). MANTIS demonstrated the highest classification accuracy (97.1%), followed by MSISensor (95.4%), and mSINGS (83.4%). MANTIS and MSISensor both exhibited equally high sensitivity (95.4%). In contrast, although mSINGS was the most specific (100%), it demonstrated poor sensitivity (66.7%), largely due to poor performance over the UCEC cohort (53.1% sensitivity). MANTIS also exhibited high specificity (98.9%), performing better than MSISensor (95.5%). To analyze disease-specific differences, results were compared between the COAD/READ and UCEC cohorts (Table 3, Supplemental Figure S3). MANTIS produced more accurate results (98.7%) than mSINGS (92.1%) and MSISensor (92.1%) in COAD/READ, whereas MSISensor was slightly more accurate (98.0%) than MANTIS (96.0%) in UCEC. While all three tools performed well with the COAD/READ data, mSINGS performed poorly with the UCEC data (accuracy 76.8%, sensitivity 53.1%). The tests showed that MANTIS had the most consistently accurate performance among the test cohort, exhibiting high sensitivity (95.4%) and specificity (98.9%) among the tested samples. Consideration of data overfitting and bias To further evaluate tool performance and to ensure MANTIS was not overfit to the COAD/READ and UCEC cohorts, tool performance was evaluated using stomach adenocarcinoma (STAD) sample pairs as a blinded test cohort (Figure 2). All three tools performed well with the STAD data, with MANTIS performing the best with 100% accuracy (Table 3), followed by MSISensor with 99% accuracy and mSINGS with 96% accuracy (Table 3). [Figure: The schematic of the MANTIS analysis for MSI detectionMicrosatellite loci are realigned against the reference genome to account for 0/1-based indexing differences. Per-locus microsatellite length distributions are determined from the normal and tumor BAM files by extracting locus-spanning reads; filtering out reads that fail to meet minimum length and average base quality requirements; determining

Read the complete text