Enhancing the Lasso Approach for Developing a Survival Prediction Model Based on Gene Expression Data

Machine Learning Model Based on Insulin Resistance Metagenes Underpins Genetic Basis of Type 2 Diabetes

Insulin resistance (IR) is considered the precursor and the key pathophysiological mechanism of type 2 diabetes (T2D) and metabolic syndrome (MetS). However, the pathways that IR shares with T2D are not clearly understood. Meta-analysis of multiple DNA microarray datasets could provide a robust set of metagenes identified across multiple studies. These metagenes would likely include a subset of genes (key metagenes) shared by both IR and T2D, and possibly responsible for the transition between them. In this study, we attempted to find these key metagenes using a feature selection method, LASSO, and then used the expression profiles of these genes to train five machine learning models: LASSO, SVM, XGBoost, Random Forest, and ANN. Among them, ANN performed well, with an area under the curve (AUC) > 95%. It also demonstrated fairly good performance in differentiating diabetics from normal glucose tolerant (NGT) persons in the test dataset, with 73% accuracy across 64 human adipose tissue samples. Furthermore, these core metagenes were also enriched in diabetes-associated terms and were found in previous genome-wide association studies of T2D and its associated glycemic traits HOMA-IR and HOMA-B. Therefore, this metagenome deserves further investigation with regard to the cardinal molecular pathological defects/pathways underlying both IR and T2D.

Predicting recurrence and recurrence-free survival in high-grade endometrial cancer using machine learning

OBJECTIVE

To develop machine-learning models to predict recurrence and time-to-recurrence in high-grade endometrial cancer (HGEC) following surgery and tailored adjuvant treatment.

METHODS

Data were retrospectively collected across eight Canadian centers including 1237 patients. Four models were trained to predict recurrence: random forests, boosted trees, and two neural networks. Receiver operating characteristic curves were used to select the best model based on the highest area under the curve (AUC). For time to recurrence, we compared random forests and Least Absolute Shrinkage and Selection Operator (LASSO) model to Cox proportional hazards.

RESULTS

The random forest was the best model to predict recurrence in HGEC; the AUCs were 85.2%, 74.1%, and 71.8% in the training, validation, and test sets, respectively. The top five predictors were: stage, uterus height, specimen weight, adjuvant chemotherapy, and preoperative histology. Performance increased to 77% and 80% when stratified by Stage III and IV, respectively. For time to recurrence, there was no difference between the LASSO and Cox proportional hazards models (c-index 71%). The random forest had a c-index of 60.5%.

CONCLUSIONS

A bootstrap random forest model may be a more accurate technique to predict recurrence in HGEC using multiple clinicopathologic factors. For time to recurrence, machine-learning methods performed similarly to the Cox proportional hazards model.

Metabolic-related gene signature model forecasts biochemical relapse in primary prostate cancer

Metabolism plays an important role in the pathogenesis of prostate cancer (PCa). Hence, we explored candidate metabolic-related genes attributed to biochemical relapse (BCR) of PCa. Gene expression profile and clinical parameters were downloaded from GSE70769 as a "training set". Using univariate Cox and LASSO-COX regression models, risk scores (RSs) were constructed. Kaplan-Meier (K-M) survival and time-dependent receiver operating characteristic (t-ROC) curves were employed. Univariate and multivariate Cox models were utilized to validate prognostic factors for biochemical relapse-free survival (BCRFS). Nomogram was plotted to facilitate clinical application. The dataset obtained from GSE70768 served as "validation set". RSs were constructed by using 7 metabolic-related genes. RSs could significantly predict 1, 3, 5-year BCRFS (AUCs for training set: 0.810-0.836; AUC for validation set: 0.673-0.827). Nomograms could effectively predicted BCRFS (training set: C-index=0.831; validation set: C-index=0.737). RSs model is an independent prognostic factor for BCR, holding greater predictive value than traditional clinicopathological parameters. Clinical Relevance- We built the prognostic nomogram based on metabolic-related gene signatures and clinicopathological features. The nomogram might further optimize biochemical relapse risk stratification for prostate cancer patients with crucial accuracy.

SurvNAM: The machine learning survival model explanation

An extension of the Neural Additive Model (NAM) called SurvNAM and its modifications are proposed to explain predictions of a black-box machine learning survival model. The method is based on applying the original NAM to solving the explanation problem in the framework of survival analysis. The basic idea behind SurvNAM is to train the network by means of a specific expected loss function which takes into account peculiarities of the survival model predictions. Moreover, the loss function approximates the black-box model by the extension of the Cox proportional hazards model, which uses the well-known Generalized Additive Model (GAM) in place of the simple linear relationship of covariates. The proposed method SurvNAM allows performing local and global explanations. The global explanation uses the whole training dataset. In contrast to the global explanation, a set of synthetic examples around the explained example are randomly generated for the local explanation. The proposed modifications of SurvNAM are based on using the Lasso-based regularization for functions from GAM and for a special representation of the GAM functions using their weighted linear and non-linear parts, which is implemented as a shortcut connection. Many numerical experiments illustrate efficiency of SurvNAM.

Bioinformatic analysis reveals an exosomal miRNA-mRNA network in colorectal cancer

BACKGROUND

Exosomes play important roles in angiogenesis, drug resistance, and metastasis of colorectal cancer (CRC), but the underlying mechanism has seldom been reported. Herein, our study aimed to reveal an exosomal miRNA-mRNA network involved in CRC by performing bioinformatical analysis.

METHODS

The mRNA and miRNA data of colon adenocarcinoma and rectal adenocarcinoma were downloaded from The Cancer Genome Atlas (TCGA) database, and exosomal miRNAs data were downloaded from the GEO dataset GSE39833. The differential expression analysis was performed using "limma" and "edgeR". Target mRNAs of miRNAs were predicted using FunRich 3.1.3, miRNAtap and multiMiR. The candidate mRNAs and exosomal miRNAs were obtained by intersecting two groups of differentially expressed miRNAs and intersection of the differential expressed mRNAs and the target mRNAs, respectively. Key mRNAs and exosomal miRNAs were identified by the least absolute shrinkage and selection operator regression analysis, and used to construct the exosomal miRNA-mRNA network. The network verified was by receiver operating characteristic curve, GEPIA and LinkedOmics. Functional enrichment analysis was also performed for studied miRNAs and mRNAs.

RESULTS

A total of 6568 differentially expressed mRNAs and 531 differentially expressed miRNAs from TCGA data, and 166 differentially expressed exosomal miRNAs in GSE39833 dataset were identified. Next, 16 key mRNAs and five key exosomal miRNAs were identified from the 5284 candidate mRNAs and 61 candidate exosomal miRNAs, respectively. The exosomal miRNA-mRNA network with high connectivity contained 13 hub mRNAs (CBFB, CDH3, ETV4, FOXQ1, FUT1, GCNT2, GRIN2D, KIAA1549, KRT80, LZTS1, SLC39A10, SPTBN2, and ZSWIM4) and five hub exosomal miRNAs (hsa-miR-126, hsa-miR-139, hsa-miR-141, hsa-miR-29c, and hsa-miR-423). The functional annotation revealed that these hub mRNAs were mainly involved in the regulation of B cell receptor signaling pathway and glycosphingolipid biosynthesis related pathways. All hub mRNAs and hub exosomal miRNAs exhibited high diagnosis value for CRC. Furthermore, the association of the hub mRNAs with overall survival, stages, and MSI phenotype of CRC revealed their important roles in CRC progression.

CONCLUSION

This study constructed an exosomal miRNA-mRNA network which may play crucial roles in the carcinogenesis and progression of CRC, thus providing potential diagnostic biomarkers and therapeutic targets for CRC.

A 10-gene signature as a predictor of biochemical recurrence after radical prostatectomy in patients with prostate cancer and a Gleason score ≥7

The time and speed of biochemical recurrence (BCR) of prostate cancer (PCa) after radical prostatectomy (RP) is highly variable. Stratification methods based on TNM staging and Gleason score (GS) do not allow the identification of patients at risk of BCR following RP. Therefore, the aim of the present study was to identify molecular signatures that can predict BCR risk effectively and facilitate treatment-related decisions for patients with PCa. RNA sequencing data and corresponding clinical data were downloaded from The Cancer Genome Atlas (TCGA) and Oncomine databases. Bioinformatics analysis was performed to identify differentially expressed genes in patients with GS=6 and GS ≥7. Cox regression models were used to determine the PCa signature (PCasig) and a clinical nomogram for the prediction of BCR. The performance of nomograms was assessed using time-dependent receiver operating characteristic curves and the concordance index (C-index). A PCasig comprising 10 genes, including SEMG2, KCNJ16, TFAP2B, SYCE1, KCNU1, AFP, GUCY1B2, GRIA4, NXPH1 and SOX11, was significantly associated with BCR, which was identified in TCGA cohort [hazard ratio (HR), 5.18; 95% CI, 3.241-8.272; C-index, 0.777] and validated in the Oncomine cohort (HR, 2.78; 95% CI, 1.39-5.54; C-index, 0.66). The expression levels of SEMG2, KCNJ16 and TFAP2B were downregulated in patients with GS ≥7. The expression levels of SYCE1, KCNU1, AFP, GUCY1B2, GRIA4, NXPH1 and SOX11 were upregulated in patients with GS ≥7. The clinical nomogram was constructed based on the GS and pathologic T stage (HR, 4.15; 95% CI, 1.39-5.54; C-index, 0.713). The addition of the PCasig to the clinical nomogram significantly improved prognostic value (HR, 7.25; 95% CI, 4.54-11.56; C-index, 0.782) with an net reclassification improvement of 75.3% (95% CI, 46.8-104.6%). Furthermore, the endogenous expression of each gene in the PCasig was measured in five PCa cell lines and in normal prostate cells, and these genes exhibited different expression levels relative to one another. In conclusion, an PCasig was identified by mining TCGA and successfully validated in an Oncomine cohort. This PCasig was an independent prognostic factor with a greater prognostic value for all patients regardless of GS than traditional clinical variables, which can improve the performance of clinical nomograms in predicting BCR of patients with GS ≥7.

TACCO, a Database Connecting Transcriptome Alterations, Pathway Alterations and Clinical Outcomes in Cancers

Because of innumerable cancer sequencing projects, abundant transcriptome expression profiles together with survival data are available from the same patients. Although some expression signatures for prognosis or pathologic staging have been identified from these data, systematically discovering such kind of expression signatures remains a challenge. To address this, we developed TACCO (Transcriptome Alterations in CanCer Omnibus), a database for identifying differentially expressed genes and altered pathways in cancer. TACCO also reveals miRNA cooperative regulations and supports construction of models for prognosis. The resulting signatures have great potential for patient stratification and treatment decision-making in future clinical applications. TACCO is freely available at http://tacco.life.nctu.edu.tw/.