Dataset Information

Systematic comparison of multi-omics survival models reveals a widespread lack of noise resistance.

ABSTRACT: As observed in several previous studies, integrating more molecular modalities in multi-omics cancer survival models may not always improve model accuracy. In this study, we compared eight deep learning and four statistical integration techniques for survival prediction on 17 multi-omics datasets, examining model performance in terms of overall accuracy and noise resistance. We found that one deep learning method, mean late fusion, and two statistical methods, PriorityLasso and BlockForest, performed best in terms of both noise resistance and overall discriminative and calibration performance. Nevertheless, all methods struggled to adequately handle noise when too many modalities were added. In summary, we confirmed that current multi-omics survival methods are not sufficiently noise resistant. We recommend relying on only modalities for which there is known predictive value for a particular cancer type until models that have stronger noise-resistance properties are developed.

SUBMITTER: Wissel D

PROVIDER: S-EPMC10162996 | biostudies-literature | 2023 Apr

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Systematic comparison of multi-omics survival models reveals a widespread lack of noise resistance.

Wissel David D Rowson Daniel D Boeva Valentina V

Cell reports methods 20230424 4

As observed in several previous studies, integrating more molecular modalities in multi-omics cancer survival models may not always improve model accuracy. In this study, we compared eight deep learning and four statistical integration techniques for survival prediction on 17 multi-omics datasets, examining model performance in terms of overall accuracy and noise resistance. We found that one deep learning method, mean late fusion, and two statistical methods, <i>PriorityLasso</i> and <i>BlockFo ...[more]

PMID: 37159669

Similar Datasets

Project description:ObjectivesPlatinum-based chemotherapies are currently the first-line treatment of non-small cell lung cancer. This study will improve our understanding of the causes of resistance to cisplatin, especially in lung adenocarcinoma (LUAD) and provide a reference for therapeutic decisions in clinical practice.MethodsCancer Cell Line Encyclopedia (CCLE), The Cancer Genome Atlas (TCGA) and Zhongshan hospital affiliated to Fudan University (zs-cohort) were used to identify the multi-omics differences related to platinum chemotherapy. Cisplatin-resistant mRNA and miRNA models were constructed by Logistic regression, classification and regression tree and C4.5 decision tree classification algorithm with previous feature selection performed via least absolute shrinkage and selection operator (LASSO). qRT-PCR and western-blotting of A549 and H358 cells, as well as single-cell Seq data of tumor samples were applied to verify the tendency of certain genes.Results661 cell lines were divided into three groups according to the IC50 value of cisplatin, and the top 1/3 (220) with a small IC50 value were defined as the sensitive group while the last 1/3 (220) were enrolled in the insensitive group. TP53 was the most common mutation in the insensitive group, in contrast to TTN in the sensitive group. 1348 mRNA, 80 miRNA, and 15 metabolites were differentially expressed between 2 groups (P < 0.05). According to the LASSO penalized logistic modeling, 6 of the 1348 mRNAs, FOXA2, BATF3, SIX1, HOXA1, ZBTB38, IRF5, were selected as the associated features with cisplatin resistance and for the contribution of predictive mRNA model (all of adjusted P-values < 0.001). Three of 6 (BATF3, IRF5, ZBTB38) genes were finally verified in cell level and patients in zs-cohort.ConclusionsSomatic mutations, mRNA expressions, miRNA expressions, metabolites and methylation were related to the resistance of cisplatin. The models we created could help in the prediction of the reaction and prognosis of patients given platinum-based chemotherapies.

Project description:Uncovering the functionally essential variations related to tumorigenesis and tumor progression from cancer genomics data is still challenging due to the genetic diversity among patients, and extensive inter- and intra-tumoral heterogeneity at different levels of gene expression regulation, including but not limited to the genomic, epigenomic, and transcriptional levels. To minimize the impact of germline genetic heterogeneities, in this study, we establish multiple primary cultures from the primary and recurrent tumors of a single patient with hepatocellular carcinoma (HCC). Multi-omics sequencing was performed for these cultures that encompass the diversity of tumor cells from the same patient. Variations in the genome sequence, epigenetic modification, and gene expression are used to infer the phylogenetic relationships of these cell cultures. We find the discrepancy among the relationships revealed by single nucleotide variations (SNVs) and transcriptional/epigenomic profiles from the cell cultures. We fail to find overlap between sample-specific mutated genes and differentially expressed genes (DEGs), suggesting that most of the heterogeneous SNVs among tumor stages or lineages of the patient are functionally insignificant. Moreover, copy number alterations (CNAs) and DNA methylation variation within gene bodies, rather than promoters, are significantly correlated with gene expression variability among these cell cultures. Pathway analysis of CNA/DNA methylation-related genes indicates that a single cell clone from the recurrent tumor exhibits distinct cellular characteristics and tumorigenicity, and such an observation is further confirmed by cellular experiments both in vitro and in vivo. Our systematic analysis reveals that CNAs and epigenomic changes, rather than SNVs, are more likely to contribute to the phenotypic diversity among subpopulations in the tumor. These findings suggest that new therapeutic strategies targeting gene dosage and epigenetic modification should be considered in personalized cancer medicine. This culture model may be applied to the further identification of plausible determinants of cancer metastasis and relapse.

Project description:BackgroundAcute myeloid leukemia (AML) is a group of highly heterogeneous diseases, for which approximately 35-40% of patients younger than 60 years old can be cured. However, the multi-omics characteristics and immune cell infiltration (ICI) status of adult long-term survival patients with AML patients compared with healthy controls are still relatively under-explored.MethodsA total of 10 healthy transplant donors (control group) and 11 long-term survival patients with AML with de novo sampling from 2019 to 2020 at the Institute of Hematology in the Hospital of Blood Diseases were enrolled. We simultaneously performed 850 K methylation and bulk RNA-seq on these 21 patients for comparing the differential gene methylation and expression levels between the two groups. The analysis of immune cell gene expression was based on 4 algorithms single sample gene set enrichment analysis (ssGSEA), EPIC, ESTIMATE and immunophenotype score (IPS) on the bulk RNA-seq data.ResultsThe differential methylation positions (DMPs) of the control group was significantly higher than that of the long-term survival group (P<0.01). The hypomethylated probes for Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment is summarized as follows: the significant pathway was related to NK-cell-mediated cytotoxicity and amino acid metabolism. We also found the Differential expression genes (DEGs) of long-term survival AML were roughly similar, and the DEGs were highly relevant to the cellular amino acid metabolic process pathway by gene set enrichment analysis (GSEA). Based on the further univariate and multivariate Cox survival analyses in GSE37642, genes crosslinked of DEGs and DMPs: LOXL1 and PDZRN4, which characterized as hypomethylated and upregulated, may become an AML prognostic marker (P<0.05). Besides, compared with the long-term-survival AML patients who discontinued chemotherapy after >3 years and the healthy donors, T cell-, natural killer cell-, MHC- and effector cell (EC)-related genes were downregulated; suppressor cells (SC) and checkpoint (CP) cells were significantly upregulated in the long-term-survival AML patients who discontinued chemotherapy after <3 years.ConclusionsIn terms of DNA methylation, RNA expression and ICI, AML patients with long-term survival were slightly different than that of healthy people. The profile of long-term AML survivors, especially those who discontinued chemotherapy less than 3 years, still differed from that of healthy people.

Dataset Information

Systematic comparison of multi-omics survival models reveals a widespread lack of noise resistance.

Publications

Systematic comparison of multi-omics survival models reveals a widespread lack of noise resistance.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets