Theory of adaptive SVD regularization for deep neural networks

Transfer Learning With Singular Value Decomposition of Multichannel Convolution Matrices

The task of transfer learning using pretrained convolutional neural networks is considered. We propose a convolution-SVD layer to analyze the convolution operators with a singular value decomposition computed in the Fourier domain. Singular vectors extracted from the source domain are transferred to the target domain, whereas the singular values are fine-tuned with a target data set. In this way, dimension reduction is achieved to avoid overfitting, while some flexibility to fine-tune the convolution kernels is maintained. We extend an existing convolution kernel reconstruction algorithm to allow for a reconstruction from an arbitrary set of learned singular values. A generalization bound for a single convolution-SVD layer is devised to show the consistency between training and testing errors. We further introduce a notion of transfer learning gap. We prove that the testing error for a single convolution-SVD layer is bounded in terms of the gap, which motivates us to develop a regularization model with the gap as the regularizer. Numerical experiments are conducted to demonstrate the superiority of the proposed model in solving classification problems and the influence of various parameters. In particular, the regularization is shown to yield a significantly higher prediction accuracy.

Prognostic and immunological role of sulfatide-related lncRNAs in hepatocellular carcinoma

Background

Hepatocellular carcinoma (HCC) is the most common primary malignancy of the liver. Long non-coding RNAs (lncRNAs) play important roles in the occurrence and development of HCC through multiple pathways. Our previous study reported the specific molecular mechanism for sulfatide regulation of integrin αV expression and cell adhesion in HCC cells through lncRNA AY927503. Next, it is necessary to identify more sulfatide-related lncRNAs, explore their clinical signifcance, and determine new targeted treatment strategies.

Methods

Microarrays were used to screen a complete set of lncRNAs with different expression profiles in sulfatide-treated cells. Sulfatide-related lncRNAs expression data and corresponding HCC patient survival information were obtained from the The Cancer Genome Atlas (TCGA) database, and the prognosis prediction model was constructed based on Cox regression analysis. Methylated RNA immunoprecipitation with next generation sequencing (MeRIP-seq) was used to detemine the effect of sulfatide on lncRNAs m6A modification. Tumor Immune Estimation Resource (TIMER) and Gene set nnrichment analysis (GSEA) were utilized to enrich the immune and functional pathways of sulfatide-related lncRNAs.

Results

A total of 85 differentially expressed lncRNAs (|Fold Change (FC)|>2, P<0.05) were screened in sulfatide-treated HCC cells. As a result, 24 sulfatide-related lncRNAs were highly expressed in HCC tissues, six of which were associated with poor prognosis in HCC patients. Based on thses data, a sulfatide-related lncRNAs prognosis assessment model for HCC was constructed. According to this risk score analysis, the overall survival (OS) curve showed that the OS of high-risk patients was significantly lower than that of low-risk patients (P<0.05). Notably, the expression difference in sulfatide-related lncRNA NRSN2-AS1 may be related to sulfatide-induced RNA m6A methylation. In addition, the expression level of NRSN2-AS1 was significantly positively correlated with immune cell infiltration in HCC and participated in the peroxisome and Peroxisome proliferator-activated receptor (PPAR) signaling pathways.

Conclusions

In conclusion, sulfatide-related lncRNAs might be promising prognostic and therapeutic targets for HCC.

Demystifying uncertainty in PM10 susceptibility mapping using variable drop-off in extreme-gradient boosting (XGB) and random forest (RF) algorithms

This study investigates uncertainty in machine learning that can occur when there is significant variance in the prediction importance level of the independent variables, especially when the ROC fails to reflect the unbalanced effect of prediction variables. A variable drop-off loop function, based on the concept of early termination for reduction of model capacity, regularization, and generalization control, was tested. A susceptibility index for airborne particulate matter of less than 10 μm diameter (PM10) was modeled using monthly maximum values and spectral bands and indices from Landsat 8 imagery, and Open Street Maps were used to prepare a range of independent variables. Probability and classification index maps were prepared using extreme-gradient boosting (XGBOOST) and random forest (RF) algorithms. These were assessed against utility criteria such as a confusion matrix of overall accuracy, quantity of variables, processing delay, degree of overfitting, importance distribution, and area under the receiver operating characteristic curve (ROC).

Identification of Prognostic Glycolysis-Related lncRNA Signature in Tumor Immune Microenvironment of Hepatocellular Carcinoma

Purpose: The purpose of this study was to construct a novel risk scoring model with prognostic value that could elucidate tumor immune microenvironment of hepatocellular carcinoma (HCC).

Samples and methods: Data were obtained through The Cancer Genome Atlas (TCGA) database. Univariate Cox analysis, least absolute shrinkage and selection operator (LASSO) analysis, and multivariate Cox analysis were carried out to screen for glycolysis-related long noncoding RNAs (lncRNAs) that could provide prognostic value. Finally, we established a risk score model to describe the characteristics of the model and verify its prediction accuracy. The receiver operating characteristic (ROC) curves of 1, 3, and 5 years of overall survival (OS) were depicted with risk score and some clinical features. ESTIMATE algorithm, single-sample gene set enrichment analysis (ssGSEA), and CIBERSORT analysis were employed to reveal the characteristics of tumor immune microenvironment in HCC. The nomogram was drawn by screening indicators with high prognostic accuracy. The correlation of risk signature with immune infiltration and immune checkpoint blockade (ICB) therapy was analyzed. After enrichment of related genes, active behaviors and pathways in high-risk groups were identified and lncRNAs related to poor prognosis were validated in vitro. Finally, the impact of MIR4435-2HG upon ICB treatment was uncovered.

Results: After screening through multiple steps, four glycolysis-related lncRNAs were obtained. The risk score constructed with the four lncRNAs was found to significantly correlate with prognosis of samples. From the ROC curve of samples with 1, 3, and 5 years of OS, two indicators were identified with high prognostic accuracy and were used to draw a nomogram. Besides, the risk score significantly correlated with immune score, immune-related signature, infiltrating immune cells (i.e. B cells, etc.), and ICB key molecules (i.e. CTLA4,etc.). Gene enrichment analysis indicated that multiple biological behaviors and pathways were active in the high-risk group. In vitro validation results showed that MIR4435-2HG was highly expressed in the two cell lines, which had a significant impact on the OS of samples. Finally, we corroborated that MIR4435-2HG had intimate relationship with ICB therapy in hepatocellular carcinoma.

Conclusion: We elucidated the crucial role of risk signature in immune cell infiltration and immunotherapy, which might contribute to clinical strategies and clinical outcome prediction of HCC.