CDMPred: A Tool for Predicting Cancer Driver Mutations with High-Quality Passenger Mutations and Ensemble Learning

doi:10.21203/rs.3.rs-1350438/v1

Download PDF

Research Article

CDMPred: A Tool for Predicting Cancer Driver Mutations with High-Quality Passenger Mutations and Ensemble Learning

https://doi.org/10.21203/rs.3.rs-1350438/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background

Benchmark datasets are of great importance for model construction and must be representative of the phenomenon being investigated. Currently available computational methods for predicting driver mutations have been trained with positive sample data, typically derived from cancer-related databases such as the Catalogue of Somatic Mutations in Cancer (COSMIC) database, whereas negative samples are generally based on statistical methods or putative samples. Obviously, negative samples are not enough for modelling the broad spectrum of passenger mutations that occur.

Results

To address these problems, we constructed a balanced dataset with driver mutations from the COSMIC database and high-quality passenger mutations from the Cancer Passenger Mutation database (dbCPM). We then encoded the features of these mutations. According to feature correlation analysis, we constructed a cancer driver missense mutations predictor (CDMPred) with feature selection via the Extreme Gradient Boosting (XGBoost) classifier ensemble learning method. The proposed CDMPred method, with the top 10 features and XGBoost, achieved area under the receiver operating characteristic curve values of 0.83 and 0.80 by 10-fold cross-validation of the training and independent test sets, respectively. Moreover, CDMPred outperformed several state-of-the-art mutation prediction methods for cancer-specific and general diseases in terms of AUC and area under the precision-recall curve.

Conclusions

High-quality passenger mutations of training data and ensemble learning are benefit to CDMPred. We expect that CDMPred can serve as a valuable method for prediction of cancer driver mutations which can further lead to deeper understanding on personal therapy.

Cancer

Machine learning

Driver mutation prediction

Benchmark quality

XGBoost

Cancer is a leading cause of death and suffering in humans, with an estimated 19.3 million new cancer diagnoses worldwide in 2020. Cancer is a complex genetic disease caused by many accumulated mutations in the genome [1]. However, only a few key mutations (drivers) contribute to the tumorigenesis and development of most cancers [2], together with certain neutral mutations (passengers). The identification of driver mutations from passenger mutations is of great significance as drivers are generally used as diagnostic and prognostic biomarkers, or as drug targets for cancer treatment.

To date, several computational methods have been designed for predicting driver mutations [3–8], including Cancer Specific High-throughput Annotation of Somatic Mutations (CHASM) [3], Cancer Driver Annotation (CanDrA) [5], and Functional Analysis through Hidden Markov Models (FATHMM) [6]. Positive samples are usually derived from cancer-related databases, such as the Catalogue of Somatic Mutations in Cancer (COSMIC) database [3–6, 8], while negative samples are generally based on statistical methods [3] or putative samples [4–6, 8].

In this paper, we assessed whether more accurate driver prediction can be attained by investigating high-quality passenger mutations. We proposed a cancer driver missense mutations predictor (CDMPred), with high-quality passenger mutations and the Extreme Gradient Boosting (XGBoost) algorithm. First, we first conducted comparative analyses of the Cancer Passenger Mutations database (dbCPM), which consists of highly curated passenger mutations [9]. Results revealed that data from the dbCPM are consistent with other negative datasets in most classical features while specific for cancer-related features [9, 10]. We then used the high-quality passenger mutation data for model training and encoded 65 features using SNVBox. Second, we compared several machine learning algorithms, based upon which we selected XGBoost. We used feature importance to investigate the correlations among the above 65 features. Finally, the optimal model (CDMPred) was leveraged with the XGBoost classifier and the top 10 features. The CDMPred method obtained AUC (area under the receiver operating characteristic (ROC) curve) values of 0.83 and 0.80 for the training and independent test sets, respectively. In addition, CDMPred outperformed several state-of-the-art mutation prediction methods for cancer-specific and general diseases based on two threshold-independent metrics (i.e., AUC and area under the precision-recall curve (AUPR)).

Analysis of features between different datasets

For all datasets shown in Additional file 1: Table S1, we quantified 85 features using dockerized CRAVAT (v5.2.3). The features are qualified in the main for a post that comprehensively represented the biological impacts of the mutation occurred in human genome [10]. With these features, the dbCPM samples were interpreted from a statistical perspective using the non-parametric Wilcoxon signed-rank hypothesis test. Significant features (P < 0.05) of positive samples sourced from DoCM and samples from dbCPM as well as other negative samples are depicted in Fig. 1A, and significant features among all negative samples are shown in Fig. 1B. We found that dbCPM data were almost identical to other negative samples in most classical features, such as ‘ExonConservation’ (conservation score for entire exon calculated from 46-species phylogenetic alignment) and ‘PredBFactorS’ (probability that wild-type residue backbone is stiff) [11, 12]. We then selected three features based on P-values, with the RMS score distribution of all samples shown in Fig. 1C. Consequently, the mutations in dbCPM were used as qualified negative samples for predicting disease-causing mutations. In accordance with expectations, dbCPM tended to show differences in cancer-specific features with other negative samples, such as ‘UniprotMETAL’ (binding site for a metal ion) and ‘UniprotREP’ (positions of repeated sequence motifs or domains) [13, 14]. The RMS score distribution of all samples for the UniprotMETAL feature is shown in Fig. 1D. These results reinforced that dbCPM mutations are more representative than other negative samples for modelling the broad spectrum of passenger mutations and more suitable for cancer driver mutation prediction.

Explorations for optimal model

We comprehensively evaluated the performance of several algorithms [15]. Eight classifiers were selected, i.e., random forest (RF), support vector machine (SVM), multi-layer perceptron (MLP), gradient boosting decision tree (GBDT), linear discriminant analysis (LDA), logistic regression (LR), naïve Bayes (NB), and XGBoost, with all algorithms implemented via scikit-learn API (v0.22.2). The classifiers were implemented with parameters optimized by grid search. The AUC training set results for the eight classifiers are shown in Fig. 2. XGBoost obviously outperformed all other classifiers, with an AUC value of 0.82. The three optimized parameters, learning_rate = 0.04, max_depth = 4, and colsample_bytree = 0.2, were applied to XGBoost.

To investigate whether the features selected from mutual information could be further refined, we determined correlations among the 65 features and found several features in UniProt that were highly related, as shown in yellow in Additional file 1: Figure S1. Consequently, we applied the feature selection method with XGBoost (default parameters) to obtain feature importance. Using sequential feature selection (SFS), we adopted the optimized parameters of XGBoost to train the data. A comparison of the AUC results for these features is illustrated in Fig. 3. The highest mean AUC (with 10-fold cross-validation) was 0.83 with the top 10 features (highlighted in bold in Additional file 1: Table S2). Among these 10 features, we compared the performance between all top 10 features and without top k (1 to 10) feature by 10-fold cross-validation (Fig. 4). Thus, we selected XGBoost with the top 10 features and optimal parameters for our final CDMPred model.

Performance comparison with state-of-the-art predictors

To assess CDMPred on unseen samples, we collected an independent test set (Table 1). We adopted several representative tools designed for cancer-specific and general disease, i.e., CHASM, CanDrA, FATHMM, TransFIC, and CScape-somatic [3, 8, 16–18]. In addition, 10 general disease predictors were collected, i.e., SIFT [19], Mutation Assessor [20], PolyPhen-2 [21], CADD [22], MetaLR [23], MetaSVM [23], DANN [24], REVEL [25], M-CAP [26], and MVP [27]. For the cancer-specific methods, we uploaded the test data to each tool website to obtain the prediction results. For the general disease predictors, we downloaded dbNSFP4.1a software (https://sites.google.com/site/jpopgen/dbNSFP) and ran the java script to retrieve prediction results from the database. All comparisons ignored the missing values of each tool. The ROC and PR curves are presented in Fig. 5 and Fig. 6, respectively. Results showed that CDMPred achieved the best performance in terms of both AUROC and AUPR.

Table 1

Summary of mutation datasets used for model construction and evaluation
	Training set		Independent test set
	Positive	Negative	Positive	Negative
Source	CHASM v3.1	dbCPM v1.0	DoCM v3.2 nonoverlap	dbCPM v1.1 nonoverlap
Number	2151	1104	567	530

Notes: DoCM v3.2 nonoverlap, data in DoCM v3.2 but not in CHASM v3.1; dbCPM v1.1 nonoverlap, data in dbCPM v1.1 except for dbCPM v1.0.

For cancer-specific methods, TransFIC (applied to PolyPhen-2 predictions due to the fewest missing values) achieved the second-best performance for AUC but the fifth-best performance for AUPR. CHASM prediction resulted in an AUC of 0.61, sensitivity of 0.74, and specificity of 0.15. Similarly, CanDrA only achieved an AUC of 0.51, sensitivity of 0.76, and specificity of 0.07. Thus, CHASM and CanDrA both showed poor performance on the negative samples, with severe imbalance yielding extremely low AUC values (as discussed below).

The CHASM training set consisted of a balanced set of positive and negative samples, however, only 0.6% overlapped at the transcript level [3]. Thus, we speculated that CHASM may be confounded by type 2 circularity, or variant status was largely predicted based on other variants in the same protein [28]. As expected, 53% of false negatives in the CHASM predictions were on transcripts that fully overlapped with positive data in the CHASM training set, while only 0.9% were on transcripts that fully overlapped with negative data in the CHASM training set. Furthermore, the reverse was found for the true negatives of the CHASM predictions, i.e., more samples were located in transcripts that exclusively overlapped with negative data in the CHASM training set. Thus, CHASM was affected by type 2 circularity.

CanDrA proposed that the driver mutations occurred recurrently in proximity (hotspots) in different types of cancer while the passenger mutations were always not in any Cancer Gene Census (CGC) genes [16, 29]. Based on our results, we suspected type 2 circularity in CanDrA as it followed such the training set screening criteria, which inevitably leads to little overlap between positive and negative samples at the transcript level. When negative sample genes in the independent test set overlapped with CGC genes, we found common genes in the two sets, which were impossible the genes for negative samples in CanDrA training set, contained 95% negative samples in the independent test set, of which only 3% were true negatives. In addition, the genes included exclusively in the negative sample genes in the independent test set, which were possible the genes for negative samples in CanDrA training set, contained 5% negative samples in the independent test set, of which > 80% were predicted as true negatives. Thus, CanDrA predicted the status of a variant based on other variants in the same protein, i.e., type 2 circularity. We demonstrated that the low AUC values acquired by both CHASM and CanDrA should be primarily attributed to type 2 circularity. In addition, from the perspective of training data quality, we propose that negative samples used in CHASM and CanDrA cannot model the broad spectrum of passenger mutations.

For the general-disease deleterious mutation predictors, CDMPred achieved the best comprehensive predictive ability, followed by CADD, Polyphen-2, and REVEL. Interestingly, these methods also outperformed the second-best cancer-specific predictor. PolyPhen-2 achieved an AUPR of 0.75 and sensitivity of 0.83, indicating relatively high predictive ability compared with CDMPred on positive samples. However, regardless of the positive or negative samples of the independent test set, many predictions in PolyPhen-2 were "positive", which may correspond to a collection of diseases, not just cancer drivers. [30]. For example, one of the true positives predicted by PolyPhen-2, “GATA2:p.R398W”, is not only related to acute myeloid leukemia, but also to alveolar proteinosis [31, 32]. In addition, one of the false negatives predicted by PolyPhen-2, “HMBS:p.D359N”, is not only related to cancer, but also to acute intermittent porphyria [33, 34]. Therefore, we focused on corresponding genes in the true negative and positive categories and false negative and positive categories of the PolyPhen-2 predictions. Enrichment analysis was performed using the online tool DAVID to verify the above suppositions [35]. We collected the pathways that only related to general diseases, not cancer, and meanwhile that an adjusted P-value < 0.05 as calculated by hypergeometric test followed by the Benjamini-Hochberg. After mapping the enrichment results at the mutation level, 65% relevant to disease of all the true negatives and true positives in PolyPhen-2 predictions, in contrast, 54% relevant to disease of all the false negatives and false positives in PolyPhen-2 predictions. In conclusion, such results supported that PolyPhen-2, even the general-disease predictors, showed a systematic bias on driver mutation prediction.

The predictive ability of machine learning methods is highly dependent on training data quality. Thus, both positive and negative samples should be consisted of known cases, and the instances were reliably defined. In this work, we proposed a novel predictor, CDMPred, for discriminating cancer missense driver mutations from passenger mutations. Here, high-quality passenger mutations from dbCPM, which were more representative than other negative samples for modelling the broad spectrum of passenger mutations, were employed as negative samples for the training set. Our results showed that high-quality passenger mutations with ensemble learning method improved upon algorithms for predicting driver mutations in human cancer. In the future, we will collect more experimentally verified negative samples and use ensemble deep learning methods to further improve the model.

The flowchart of the proposed CDMPred method is provided in Fig. 7. In detail, the procedure is composed of four steps, i.e., dataset preparation, feature representation, model construction, and performance evaluation, which are elucidated as follows.

Dataset preparation

The datasets were divided into two groups for feature analysis and model construction, respectively. The dbCPM contains 1 919 passenger mutations, with 1 634 distinct missense mutations [9]. To consider the reliability of negative samples, we statistically compared the data in dbCPM with gold-standard positive samples and traditional negative samples. For positive samples, 1 248 driver missense mutations were taken from the Database of Curated Mutations (DoCM) (v3.2) [36], which aggregates functionally validated mutations in cancer with high reliability. For negative samples, we collected two datasets composed of training sets from classic prediction tools for cancer-specific driver mutations, i.e., all negative samples in the FATHMM training set (originally obtained from the UniProt database) [17, 37] and all synthetic passenger mutations in the CHASM (v3.1) training set [3]. All data used for comparative analysis are shown in Additional file 1: Table S1.

The datasets used for model construction are described as follows. Of the 1 634 missense passenger mutations in dbCPM, 1 104 items in our training set were reported in the database before first updated. The 13 235 positive samples in the CHASM (v3.1) training set were filtered. Considering the balance of positive and negative samples in the training set, we only adopted positive samples located within 50 bp of a negative mutation on the same transcript. As a result, 2 151 driver missense mutations were used in our training set. For performance comparison with state-of-the-art prediction tools, we also obtained an independent test set. Firstly, of the 1 634 missense mutations in dbCPM, only the items that reported after first updated in the database were collected in the independent test set. Secondly, we considered all 1 248 driver mutations in DoCM as the positive sample source. Finally, we removed data that overlapped with the training set to avoid model confounding by type 1 circularity [28], which can occur due to overlap between training and evaluation datasets, resulting in potential overfitting. The datasets used for model construction and performance evaluation are shown in Table 1.

Feature representation

For each missense mutation in the above dataset, 85 features were obtained from the Cancer-Related Analysis of Variants Toolkit (CRAVAT (v5.2.3), https://hub.docker.com/r/karchinlab/cravatmupit/) [38] in Linux. CRAVAT is an evolving suite of informatics tools for genomic variant interpretation. To prepare the input data, we curated the transcript information using Ensembl GRCh37 [39]. The 85 features encoded included evolutionary conservation scores, protein physicochemical properties, sequence context indices, and UniProtKB annotations. Each feature was scaled by subtracting the mean value and dividing by the root-mean-square (RMS) value using all values precomputed for the whole genome. Following CHASM, we implemented the information gain method on the training set [19, 20]. After setting the same threshold to retain only features with at least 0.001 bits of mutual information, 65 predictive features were selected, as shown in Additional file 1: Table S2.

Model construction

We used feature importance with XGBoost to select an optimal feature subset. Compared with several machine learning methods, the ultimate CDMPred model was based on XGBoost, which is a scalable and flexible gradient boosting method [40]. Here, the “XGBClassifier” package was employed and implemented with 10-fold cross-validation. In addition, we optimized three major parameters, i.e., boosting learning rate (learning_rate), maximum depth of tree (maxDepth), and subsample ratio of columns when constructing each tree (colsample_bytree), using a grid search based on the 10-fold cross-validation results of the training set.

Performance evaluation

We used two threshold-independent measures, i.e., AUC and AUPR, as quantitative measurements of prediction results [3, 23]. We also adopted two qualitative measures, i.e., sensitivity (or true positive rate) and specificity (true negative rate), for model performance analysis, as described in previous research [41, 42] and defined as follows:

where TP (true positive) means the number of correctly predicted cancer driver mutations, FP (false positive) represents the number of passenger mutations predicted as drivers, TN (true negative) represents the number of correctly predicted passenger mutations, and FN (false negative) indicates the number of cancer driver mutations predicted as passengers.

COSMIC

the Catalogue of Somatic Mutations in Cancer

dbCPM

the Cancer Passenger Mutation database

XGBoost

Extreme Gradient Boosting

AUC

area under the receiver operating characteristic curve

CHASM

Cancer Specific High-throughput Annotation of Somatic Mutations

CanDrA

Cancer Driver Annotation

FATHMM

Functional Analysis through Hidden Markov Models

SFS

sequential feature selection

random forest

SVM

support vector machine

MLP

multi-layer perceptron

GBDT

gradient boosting decision tree

LDA

linear discriminant analysis

logistic regression

naïve Bayes.

Ethics approval and consent to participate

All methods were performed in accordance with the relevant guidelines and regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Availability of data and materials

The codes and data supporting the conclusions of this article can be obtained from https://github.com/xialab-ahu/CDMPred.

Funding

This work was supported by the National Natural Science Foundation of China (11835014, 62072003, 82101611, and U19A2064) and the Natural Science Young Foundation of Anhui (2008085QF293).

Authors’ contributions

X.Y.L. and J.F.X. designed the project, L.H.W and H.Y.S. collected the data, H.Y.S, L.H.W., Z.Y.Y., and J.F.X. analyzed the data, H.Y.S, L.H.W., and Z.Y.Y. provided constructive suggestions and discussions during the project, X.Y.L., L.H.W. and H.Y.S. wrote the manuscript. All authors read and approved the final manuscript.

Acknowledgements

The authors thank the members of our laboratory for their valuable contributions to CDMPred.

Wood LD, Parsons DW, Jones S, Lin J, Sjoblom T, Leary RJ, Shen D, Boca SM, Barber T, Ptak J et al: The Genomic Landscapes of Human Breast and Colorectal Cancers. Science 2007, 318(5853):1108–1113.
Hanahan D, Weinberg RA: Hallmarks of Cancer: The Next Generation. Cell 2011, 144(5):646–674.
Carter H, Chen S, Isik L, Tyekucheva S, Velculescu VE, Kinzler KW, Vogelstein B, Karchin R: Cancer-specific high-throughput annotation of somatic mutations: computational prediction of driver missense mutations. Cancer Res 2009, 69(16):6660–6667.
Gonzalez-Perez A, Deu-Pons J, Lopez-Bigas N: Improving the prediction of the functional impact of cancer mutations by baseline tolerance transformation. Genome Med 2012, 4(11):89.
Mao Y, Chen H, Liang H, Meric-Bernstam F, Mills GB, Chen K: CanDrA: cancer-specific driver missense mutation annotation with optimized features. PLoS One 2013, 8(10):e77945.
Shihab HA, Gough J, Cooper DN, Day IN, Gaunt TR: Predicting the functional consequences of cancer-associated amino acid substitutions. Bioinformatics 2013, 29(12):1504–1510.
Kumar RD, Swamidass SJ, Bose R: Unsupervised detection of cancer driver mutations with parsimony-guided learning. Nat Genet 2016, 48(10):1288–1294.
Rogers MF, Gaunt TR, Campbell C: CScape-somatic: distinguishing driver and passenger point mutations in the cancer genome. Bioinformatics 2020, 36(12):3637–3644.
Yue Z, Zhao L, Xia J: dbCPM: a manually curated database for exploring the cancer passenger mutations. Brief Bioinform 2020, 21(1):309–317.
Wong WC, Kim D, Carter H, Diekhans M, Ryan MC, Karchin R: CHASM and SNVBox: toolkit for detecting biologically important single nucleotide mutations in cancer. Bioinformatics 2011, 27(15):2147–2148.
Kent WJ, Sugnet CW, Furey TS, Roskin KM, Pringle TH, Zahler AM, Haussler D: The human genome browser at UCSC. Genome research 2002, 12(6):996–1006.
Katzman S, Barrett C, Thiltgen G, Karchin R, Karplus K: PREDICT-2ND: a tool for generalized protein local structure prediction. Bioinformatics 2008, 24(21):2453–2459.
Ribeiro GRH, Francisco G, Teixeira LVS, Romao-Correia RF, Sanches JA, Neto CF, Ruiz IRG: Repetitive DNA alterations in human skin cancers. Journal of Dermatological Science 2004, 36(2):79–86.
Xu D, Jalal SI, Sledge GW, Jr., Meroueh SO: Small-molecule binding sites to explore protein-protein interactions in the cancer proteome. Molecular Biosystems 2016, 12(10):3067–3087.
Buske OJ, Manickaraj A, Mital S, Ray PN, Brudno M: Identification of deleterious synonymous variants in human genomes. Bioinformatics 2013, 29(15):1843–1850.
Mao Y, Chen H, Liang H, Meric-Bernstam F, Mills G, Chen K: CanDrA: Cancer-specific driver missense mutation annotation with optimized features. Cancer research 2013, 73(8):e77945.
Shihab HA, Gough J, Cooper DN, Day INM, Gaunt TR: Predicting the functional consequences of cancer-associated amino acid substitutions. Bioinformatics 2013, 29(12):1504–1510.
Gonzalez-Perez A, Deu-Pons J, Lopez-Bigas N: Improving the prediction of the functional impact of cancer mutations by baseline tolerance transformation. Genome Medicine 2012, 4.
Kumar P, Henikoff S, Ng PC: Predicting the effects of coding non-synonymous variants on protein function using the SIFT algorithm. Nat Protoc 2009, 4(7):1073–1081.
Reva B, Antipin Y, Sander C: Predicting the functional impact of protein mutations: application to cancer genomics. Nucleic Acids Res 2011, 39(17):e118.
Adzhubei IA, Schmidt S, Peshkin L, Ramensky VE, Gerasimova A, Bork P, Kondrashov AS, Sunyaev SR: A method and server for predicting damaging missense mutations. Nat Methods 2010, 7(4):248–249.
Kircher M, Witten DM, Jain P, O'Roak BJ, Cooper GM, Shendure J: A general framework for estimating the relative pathogenicity of human genetic variants. Nat Genet 2014, 46(3):310–315.
Dong C, Wei P, Jian X, Gibbs R, Boerwinkle E, Wang K, Liu X: Comparison and integration of deleteriousness prediction methods for nonsynonymous SNVs in whole exome sequencing studies. Hum Mol Genet 2015, 24(8):2125–2137.
Quang D, Chen Y, Xie X: DANN: a deep learning approach for annotating the pathogenicity of genetic variants. Bioinformatics 2015, 31(5):761–763.
Ioannidis NM, Rothstein JH, Pejaver V, Middha S, McDonnell SK, Baheti S, Musolf A, Li Q, Holzinger E, Karyadi D et al: REVEL: An Ensemble Method for Predicting the Pathogenicity of Rare Missense Variants. Am J Hum Genet 2016, 99(4):877–885.
Jagadeesh KA, Wenger AM, Berger MJ, Guturu H, Stenson PD, Cooper DN, Bernstein JA, Bejerano G: M-CAP eliminates a majority of variants of uncertain significance in clinical exomes at high sensitivity. Nat Genet 2016, 48(12):1581–1586.
Qi H, Chen C, Zhang H, Long JJ, Chung WK, Guan Y, Shen Y: MVP: predicting pathogenicity of missense variants by deep neural networks. 2018.
Grimm DG, Azencott CA, Aicheler F, Gieraths U, MacArthur DG, Samocha KE, Cooper DN, Stenson PD, Daly MJ, Smoller JW et al: The evaluation of tools used to predict the impact of missense variants is hindered by two types of circularity. Hum Mutat 2015, 36(5):513–523.
Futreal PA, Coin L, Marshall M, Down T, Hubbard T, Wooster R, Rahman N, Stratton MR: A census of human cancer genes. Nature Reviews Cancer 2004, 4(3):177–183.
Bertrand D, Drissler S, Chia BK, Koh JY, Li C, Suphavilai C, Tan IB, Nagarajan N: ConsensusDriver Improves upon Individual Algorithms for Predicting Driver Alterations in Different Cancer Types and Individual Patients. Cancer research 2018, 78(1):290–301.
Kazenwadel J, Secker GA, Liu YJ, Rosenfeld JA, Wildin RS, Cuellar-Rodriguez J, Hsu AP, Dyack S, Fernandez CV, Chong C-E et al: Loss-of-function germline GATA2 mutations in patients with MDS/AML or MonoMAC syndrome and primary lymphedema reveal a key role for GATA2 in the lymphatic vasculature. Blood 2012, 119(5):1283–1291.
Griese M, Zarbock R, Costabel U, Hildebrandt J, Theegarten D, Albert M, Thiel A, Schams A, Lange J, Krenke K et al: GATA2 deficiency in children and adults with severe pulmonary alveolar proteinosis and hematologic disorders. Bmc Pulmonary Medicine 2015, 15.
Dorschner MO, Amendola LM, Turner EH, Robertson PD, Shirts BH, Gallego CJ, Bennett RL, Jones KL, Tokita MJ, Bennett JT et al: Actionable, Pathogenic Incidental Findings in 1,000 Participants' Exomes. American Journal of Human Genetics 2013, 93(4):631–640.
Lewis PD: Novel human pathological mutations. Human Genetics 2006, 118(6):359–364.
Huang DW, Sherman BT, Lempicki RA: Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nature protocols 2009, 4(1):44–57.
Ainscough BJ, Griffith M, Coffman AC, Wagner AH, Kunisaki J, Choudhary MNK, McMichael JF, Fulton RS, Wilson RK, Griffith OL et al: DoCM: a database of curated mutations in cancer. Nature Methods 2016, 13(10):806–807.
Apweiler R, Bairoch A, Wu CH, Barker WC, Boeckmann B, Ferro S, Gasteiger E, Huang HZ, Lopez R, Magrane M et al: UniProt: the Universal Protein knowledgebase. Nucleic acids research 2004, 32:D115-D119.
Masica DL, Douville C, Tokheim C, Bhattacharya R, Kim R, Moad K, Ryan MC, Karchin R: CRAVAT 4: Cancer-Related Analysis of Variants Toolkit. Cancer Res 2017, 77(21):e35-e38.
Flicek P, Amode MR, Barrell D, Beal K, Billis K, Brent S, Carvalho-Silva D, Clapham P, Coates G, Fitzgerald S et al: Ensembl 2014. Nucleic acids research 2014, 42(D1):D749-D755.
Chen T, Guestrin C: Xgboost: A scalable tree boosting system[C]. ACM 2016:785–794.
Cheng N, Li M, Zhao L, Zhang B, Yang Y, Zheng C-H, Xia J: Comparison and integration of computational methods for deleterious synonymous mutation prediction. Brief Bioinform 2020, 21(3):970–981.
Zeng X, Liu L, Lü L, Zou Q: Prediction of potential disease-associated microRNAs using structural perturbation method. Bioinformatics 2018, 34(14):2425–2432.

No competing interests reported.

Additionalfile1.pdf

Download PDF

Version 1

posted

You are reading this latest preprint version

CDMPred: A Tool for Predicting Cancer Driver Mutations with High-Quality Passenger Mutations and Ensemble Learning

Status:

Version 1

Abstract

Background

Results

Conclusions

Figures

Background

Results

Analysis of features between different datasets

Explorations for optimal model

Performance comparison with state-of-the-art predictors

Discussion

Conclusions

Methods

Dataset preparation

Feature representation

Model construction

Performance evaluation

List Of Abbreviations

Declarations

Ethics approval and consent to participate

Consent for publication

Competing interests

Availability of data and materials

Funding

Authors’ contributions

Acknowledgements

References

Additional Declarations

Supplementary Files

Status:

Version 1