Manifold learning based robust clustering of gene expression data for cancer subtyping

ForestSubtype: a cancer subtype identifying approach based on high-dimensional genomic data and a parallel random forest

BACKGROUND

Cancer subtype classification is helpful for personalized cancer treatment. Although, some approaches have been developed to classifying caner subtype based on high dimensional gene expression data, it is difficult to obtain satisfactory classification results. Meanwhile, some cancers have been well studied and classified to some subtypes, which are adopt by most researchers. Hence, this priori knowledge is significant for further identifying new meaningful subtypes.

RESULTS

In this paper, we present a combined parallel random forest and autoencoder approach for cancer subtype identification based on high dimensional gene expression data, ForestSubtype. ForestSubtype first adopts the parallel RF and the priori knowledge of cancer subtype to train a module and extract significant candidate features. Second, ForestSubtype uses a random forest as the base module and ten parallel random forests to compute each feature weight and rank them separately. Then, the intersection of the features with the larger weights output by the ten parallel random forests is taken as our subsequent candidate features. Third, ForestSubtype uses an autoencoder to condenses the selected features into a two-dimensional data. Fourth, ForestSubtype utilizes k-means++ to obtain new cancer subtype identification results. In this paper, the breast cancer gene expression data obtained from The Cancer Genome Atlas are used for training and validation, and an independent breast cancer dataset from the Molecular Taxonomy of Breast Cancer International Consortium is used for testing. Additionally, we use two other cancer datasets for validating the generalizability of ForestSubtype. ForestSubtype outperforms the other two methods in terms of the distribution of clusters, internal and external metric results. The open-source code is available at https://github.com/lffyd/ForestSubtype .

CONCLUSIONS

Our work shows that the combination of high-dimensional gene expression data and parallel random forests and autoencoder, guided by a priori knowledge, can identify new subtypes more effectively than existing methods of cancer subtype classification.

Optimized cell type signatures revealed from single-cell data by combining principal feature analysis, mutual information, and machine learning

Machine learning techniques are excellent to analyze expression data from single cells. These techniques impact all fields ranging from cell annotation and clustering to signature identification. The presented framework evaluates gene selection sets how far they optimally separate defined phenotypes or cell groups. This innovation overcomes the present limitation to objectively and correctly identify a small gene set of high information content regarding separating phenotypes for which corresponding code scripts are provided. The small but meaningful subset of the original genes (or feature space) facilitates human interpretability of the differences of the phenotypes including those found by machine learning results and may even turn correlations between genes and phenotypes into a causal explanation. For the feature selection task, the principal feature analysis is utilized which reduces redundant information while selecting genes that carry the information for separating the phenotypes. In this context, the presented framework shows explainability of unsupervised learning as it reveals cell-type specific signatures. Apart from a Seurat preprocessing tool and the PFA script, the pipeline uses mutual information to balance accuracy and size of the gene set if desired. A validation part to evaluate the gene selection for their information content regarding the separation of the phenotypes is provided as well, binary and multiclass classification of 3 or 4 groups are studied. Results from different single-cell data are presented. In each, only about ten out of more than 30000 genes are identified as carrying the relevant information. The code is provided in a GitHub repository at https://github.com/AC-PHD/Seurat_PFA_pipeline.

Robust correlation estimation and UMAP assisted topological analysis of omics data for disease subtyping

Deciphering information hidden in the gene expression assays for identifying disease subtypes has significant importance in precision medicine. However, computational limitations thwart this process due to the intricacy of the biological networks and the curse of dimensionality of gene expression data. Therefore, clustering in such scenarios often becomes the first choice of exploratory data analysis to identify natural structures and intrinsic patterns in the data. However, sparse and high dimensional nature of omics data prevents conventional clustering algorithms to discover subtypes that are clinically relevant and statistically significant. Hence, non-linear dimensionality reduction techniques coupled with clustering in such scenarios often becomes imperative to improve the clustering results. In this study, we present a robust pipeline to discover disease subtypes with clinical relevance. Specifically, we focus on discovering patient sub-groups that have a residual life patterns remarkably different from other sub-groups. This is significant because by refining prognosis, subtyping can reduce uncertainty in approximating patients expected outcome. The methodology present is based on robust correlation estimation, UMAP- a non-linear dimensionality reduction method and mapper- a tool from topology. Notably, we suggest a method for improving the robustness of the correlation matrix of gene expression data for improving the clustering results. The performance of the model is evaluated by applying to five cancer datasets obtained through TCGA and comparisons are performed with some state of the art methods of NEMO, RSC-OTRI and SNF with regard to log-rank test and Restricted Life Expectancy Difference. For example in GBM dataset, the minimum separation for any two discovered subtypes is 221 days which is significantly higher than the other methodologies. We also compared the results without using the robust correlation based estimate and observed that robust correlation improves separability between survival curves significantly. From the results we infer that our methodology performs better compared to other methodologies with regard to separating survival curves of patient sub-groups despite using single omics profiles of patients compared to multiple omics profiles of SNF and NEMO. Pathway over-representation analysis is performed on the final clustering results to investigate the biological underpinnings characterizing each subtype.

Optimization of synthetic molecular reporters for a mesenchymal glioblastoma transcriptional program by integer programing

MOTIVATION

A recent approach to perform genetic tracing of complex biological problems involves the generation of synthetic deoxyribonucleic acid (DNA) probes that specifically mark cells with a phenotype of interest. These synthetic locus control regions (sLCRs), in turn, drive the expression of a reporter gene, such as fluorescent protein. To build functional and specific sLCRs, it is critical to accurately select multiple bona fide cis-regulatory elements from the target cell phenotype cistrome. This selection occurs by maximizing the number and diversity of transcription factors (TFs) within the sLCR, yet the size of the final sLCR should remain limited.

RESULTS

In this work, we discuss how optimization, in particular integer programing, can be used to systematically address the construction of a specific sLCR and optimize pre-defined properties of the sLCR. Our presented instance of a linear optimization problem maximizes the activation potential of the sLCR such that its size is limited to a pre-defined length and a minimum number of all TFs deemed sufficiently characteristic for the phenotype of interest is covered. We generated an sLCR to trace the mesenchymal glioblastoma program in patients by solving our corresponding linear program with the software optimizer Gurobi. Considering the binding strength of transcription factor binding sites (TFBSs) with their TFs as a proxy for activation potential, the optimized sLCR scores similarly to an sLCR experimentally validated in vivo, and is smaller in size while having the same coverage of TFBSs.

AVAILABILITY AND IMPLEMENTATION

We provide a Python implementation of the presented framework in the Supplementary Material with which an optimal selection of cis-regulatory elements can be calculated once the target set of TFs and their binding strength with their TFBSs is known.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.