ellipsoidFN: a tool for identifying a heterogeneous set of cancer biomarkers based on gene expressions

Application of artificial intelligence and machine learning for prediction of oral cancer risk

BACKGROUND

Oral cancer requires early diagnosis and treatment to increase the chances of survival. This study aimed to develop an artificial neural network model that helps to predict the individuals' risk of developing oral cancer based on data on risk factors, systematic medical condition, and clinic-pathological features.

METHODS

A popular data mining algorithm artificial neural network was used for developing the artificial intelligence-based prediction model. A total of 29 variables that were associated with the patients were used for developing the model. The dataset was randomly split into the training dataset 54 (75%) cases and testing dataset 19 (25%) cases. All records and observations were reviewed by Board-certified oral pathologist.

RESULTS

A total of 73 patients met the eligibility criteria. Twenty-two (30.13%) were benign cases, and 51 (69.86%) were malignant cases. Thirty-seven were female, and 36 were male, with a mean age of 63.09 years. Our analysis displayed that the average sensitivity and specificity of ANN for oral cancer prediction based on the 10-fold cross-validation analysis was 85.71% (95% confidence interval [CI], 57.19-98.22) and 60.00% (95% CI, 14.66-94.73), respectively. The accuracy of ANN for oral cancer prediction was 78.95% (95% CI, 54.43-931.95).

CONCLUSION

Our results suggest that this machine-learning technique has the potential to help in oral cancer screening and diagnosis based on the datasets. The results demonstrate that the artificial neural network could perform well in estimating the probability of malignancy and improve the positive predictive value that could help to predict the individuals' risk of developing OC based on knowledge of their risk factors, systemic medical conditions, and clinic-pathological data.

Iterative sure independent ranking and screening for drug response prediction

Background

Prediction of drug response based on multi-omics data is a crucial task in the research of personalized cancer therapy.

Results

We proposed an iterative sure independent ranking and screening (ISIRS) scheme to select drug response-associated features and applied it to the Cancer Cell Line Encyclopedia (CCLE) dataset. For each drug in CCLE, we incorporated multi-omics data including copy number alterations, mutation and gene expression and selected up to 50 features using ISIRS. Then a linear regression model based on the selected features was exploited to predict the drug response. Cross validation test shows that our prediction accuracies are higher than existing methods for most drugs.

Conclusions

Our study indicates that the features selected by the marginal utility measure, which measures the conditional probability of drug responses given the feature, are helpful for drug response prediction.

Computer-aided biomarker discovery for precision medicine: data resources, models and applications

Biomarkers are a class of measurable and evaluable indicators with the potential to predict disease initiation and progression. In contrast to disease-associated factors, biomarkers hold the promise to capture the changeable signatures of biological states. With methodological advances, computer-aided biomarker discovery has now become a burgeoning paradigm in the field of biomedical science. In recent years, the 'big data' term has accumulated for the systematical investigation of complex biological phenomena and promoted the flourishing of computational methods for systems-level biomarker screening. Compared with routine wet-lab experiments, bioinformatics approaches are more efficient to decode disease pathogenesis under a holistic framework, which is propitious to identify biomarkers ranging from single molecules to molecular networks for disease diagnosis, prognosis and therapy. In this review, the concept and characteristics of typical biomarker types, e.g. single molecular biomarkers, module/network biomarkers, cross-level biomarkers, etc., are explicated on the guidance of systems biology. Then, publicly available data resources together with some well-constructed biomarker databases and knowledge bases are introduced. Biomarker identification models using mathematical, network and machine learning theories are sequentially discussed. Based on network substructural and functional evidences, a novel bioinformatics model is particularly highlighted for microRNA biomarker discovery. This article aims to give deep insights into the advantages and challenges of current computational approaches for biomarker detection, and to light up the future wisdom toward precision medicine and nation-wide healthcare.

Integrating Artificial Intelligence and Nanotechnology for Precision Cancer Medicine

Artificial intelligence (AI) and nanotechnology are two fields that are instrumental in realizing the goal of precision medicine-tailoring the best treatment for each cancer patient. Recent conversion between these two fields is enabling better patient data acquisition and improved design of nanomaterials for precision cancer medicine. Diagnostic nanomaterials are used to assemble a patient-specific disease profile, which is then leveraged, through a set of therapeutic nanotechnologies, to improve the treatment outcome. However, high intratumor and interpatient heterogeneities make the rational design of diagnostic and therapeutic platforms, and analysis of their output, extremely difficult. Integration of AI approaches can bridge this gap, using pattern analysis and classification algorithms for improved diagnostic and therapeutic accuracy. Nanomedicine design also benefits from the application of AI, by optimizing material properties according to predicted interactions with the target drug, biological fluids, immune system, vasculature, and cell membranes, all affecting therapeutic efficacy. Here, fundamental concepts in AI are described and the contributions and promise of nanotechnology coupled with AI to the future of precision cancer medicine are reviewed.

A multitask multiple kernel learning formulation for discriminating early- and late-stage cancers

Motivation

Genomic information is increasingly being used in diagnosis, prognosis and treatment of cancer. The severity of the disease is usually measured by the tumor stage. Therefore, identifying pathways playing an important role in progression of the disease stage is of great interest. Given that there are similarities in the underlying mechanisms of different cancers, in addition to the considerable correlation in the genomic data, there is a need for machine learning methods that can take these aspects of genomic data into account. Furthermore, using machine learning for studying multiple cancer cohorts together with a collection of molecular pathways creates an opportunity for knowledge extraction.

Results

We studied the problem of discriminating early- and late-stage tumors of several cancers using genomic information while enforcing interpretability on the solutions. To this end, we developed a multitask multiple kernel learning (MTMKL) method with a co-clustering step based on a cutting-plane algorithm to identify the relationships between the input tasks and kernels. We tested our algorithm on 15 cancer cohorts and observed that, in most cases, MTMKL outperforms other algorithms (including random forests, support vector machine and single-task multiple kernel learning) in terms of predictive power. Using the aggregate results from multiple replications, we also derived similarity matrices between cancer cohorts, which are, in many cases, in agreement with available relationships reported in the relevant literature.

Availability and implementation

Our implementations of support vector machine and multiple kernel learning algorithms in R are available at https://github.com/arezourahimi/mtgsbc together with the scripts that replicate the reported experiments.

Supplementary information

Supplementary data are available at Bioinformatics online.

ELF: Extract Landmark Features By Optimizing Topology Maintenance, Redundancy, and Specificity

Feature selection is the process of selecting a subset of landmark features for model construction when there are many features and a comparatively few samples. The far-reaching development technologies such as biological sequencing at single cell level make feature selection a more challenging work. The difficulty lies in four facts: those features measured are in high dimension and with noise; dropouts make the data much sparse; many features are either redundant or irrelevant; and samples are not well-labeled in the experiments. Here, we propose a new model called ELF (Extract Landmark Features) to address the above challenges. ELF aims to simultaneously maximize topology maintenance to keep the pairwise relationships among samples, minimize feature redundancy to diversify the features, and maximize feature specificity to make every selected feature more representative. This makes ELF a nonlinear combinatorial optimization. To solve this difficult problem, we propose a heuristic algorithm based on greedy strategy. We show ELF's outstanding performance on two single cell RNA-seq datasets. One is the direct reprogramming from mouse embryonic fibroblasts to induced neuron and the other is hepatoblast differentiation. ELF is able to choose only hundreds of landmark genes to maintain the cells' correlativity. Topology maintenance, redundancy removal, and specificity each plays its important role in selecting landmark features and revealing cells' biological functions. In addition, ELF can be generally applied in other scenarios. We demonstrate that ELF can reveal pivotal pixel in writing region and human face in two public image datasets. We believe that ELF is a useful tool to obtain more interpretable results by revealing key features while clustering the samples well.

Identification of potential genes and pathways for response prediction of neoadjuvant chemoradiotherapy in patients with rectal cancer by systemic biological analysis

Currently, neoadjuvant chemoradiotherapy (CRT) followed by radical surgery is the standard of care for locally advanced rectal cancer. However, to the best of our knowledge, there are no effective biomarkers for predicting patients who may benefit from neoadjuvant treatment. The aim of the current study was to screen potential crucial genes and pathways associated with the response to CRT in rectal cancer, and provide valid biological information to assist further investigation of CRT optimization. In the current study, differentially expressed (DE) genes were identified from the tumor samples of responders and non-responders to neoadjuvant CRT in the GSE35452 gene expression profile. Seven hub genes and one significant module were identified from the protein-protein interaction (PPI) network. Functional enrichment analysis of all the DE genes and the hub genes, retrieved from PPI network analysis, revealed their associations with CRT response. Genes were identified that may be used to discriminate patients who would or would not clinically benefit from neoadjuvant CRT. Several important pathways enriched by the DE genes, hub genes and selected module were identified, and revealed to be closely associated with radiation response, including excision repair, homologous recombination, Ras signaling pathway, the forkhead box O signaling pathway, focal adhesion and the Wnt signaling pathway. In conclusion, the current study demonstrated that the identified gene signatures and pathways may be used as molecular biomarkers for predicting CRT response. Furthermore, combinations of these biomarkers may be helpful for optimizing CRT treatment and promoting understanding of the molecular basis of response differences; this needs to be confirmed by further experiments.

Screening candidate microRNA-mRNA regulatory pairs for predicting the response to chemoradiotherapy in rectal cancer by a bioinformatics approach

Extensive efforts have been undertaken in search of biomarkers for predicting the chemoradiotherapy response in rectal cancer. However, most attention on treatment efficiency prediction in carcinoma is addicted to single or limited molecules. Network biomarkers are considered to outperform single molecules in predictive power. In this study, candidate microRNAs (miRNAs) were identified from the PubMed citations and miRNA expression profiles. Targets of miRNAs were obtained from four experimentally confirmed interactions and three computationally predicted databases. Functional enrichment analysis of all the targets revealed their associations with chemoradiotherapy response, indicating they could be promising biomarkers. Two lists of key target mRNAs of the candidate miRNAs were retrieved from protein–protein interaction (PPI) network and mRNA expression profiles, respectively. Pathway analysis and literature validation revealed that the mRNA lists were highly related to the ionizing radiation. The above miRNAs along with the key miRNA targets provide potential miRNA-mRNA regulatory pairs as network biomarkers in which all the network components may be used for predicting the chemoradiotherapy response. These results demonstrated that the network biomarkers could provide a useful model for predicting the chemoradiotherapy response and help in further understanding the molecular basis of response differences, which should be prioritized for further study.

A New Approach for Feature Selection from Microarray Data Based on Mutual Information

Mutual information (MI) is a powerful concept for correlation-centric applications. It has been used for feature selection from microarray gene expression data in many works. One of the merits of MI is that, unlike many other heuristic methods, it is based on a mature theoretic foundation. When applied to microarray data, however, it faces some challenges. First, due to the large number of features (i.e., genes) present in microarray data, the true distributions for the expression values of some genes may be distorted by noise. Second, evaluating inter-group mutual information requires estimating multi-variate distributions, which is quite difficult if not impossible. To address these problems, in this paper, we propose a new MI-based feature selection approach for microarray data. Our approach relies on two strategies: one is relevance boosting, which requires a desirable feature to show substantially additional relevance with class labeling beyond the already selected features, the other is feature interaction enhancing, which probabilistically compensates for feature interaction missing from simple aggregation-based evaluation. We justify our approach from both theoretical perspective and experimental results. We use a synthetic dataset to show the statistical significance of the proposed strategies, and real-life datasets to show the improved performance of our approach over the existing methods.

An Efficient Approach to Screening Epigenome-Wide Data

Screening cytosine-phosphate-guanine dinucleotide (CpG) DNA methylation sites in association with some covariate(s) is desired due to high dimensionality. We incorporate surrogate variable analyses (SVAs) into (ordinary or robust) linear regressions and utilize training and testing samples for nested validation to screen CpG sites. SVA is to account for variations in the methylation not explained by the specified covariate(s) and adjust for confounding effects. To make it easier to users, this screening method is built into a user-friendly R package, ttScreening, with efficient algorithms implemented. Various simulations were implemented to examine the robustness and sensitivity of the method compared to the classical approaches controlling for multiple testing: the false discovery rates-based (FDR-based) and the Bonferroni-based methods. The proposed approach in general performs better and has the potential to control both types I and II errors. We applied ttScreening to 383,998 CpG sites in association with maternal smoking, one of the leading factors for cancer risk.

Machine learning for biomarker identification in cancer research - developments toward its clinical application

The patterns identified from the systematically collected molecular profiles of patient tumor samples, along with clinical metadata, can assist personalized treatments for effective management of cancer patients with similar molecular subtypes. There is an unmet need to develop computational algorithms for cancer diagnosis, prognosis and therapeutics that can identify complex patterns and help in classifications based on plethora of emerging cancer research outcomes in public domain. Machine learning, a branch of artificial intelligence, holds a great potential for pattern recognition in cryptic cancer datasets, as evident from recent literature survey. In this review, we focus on the current status of machine learning applications in cancer research, highlighting trends and analyzing major achievements, roadblocks and challenges toward its implementation in clinics.

Identifying network biomarkers based on protein-protein interactions and expression data

Identifying effective biomarkers to battle complex diseases is an important but challenging task in biomedical research today. Molecular data of complex diseases is increasingly abundant due to the rapid advance of high throughput technologies. However, a great gap remains in identifying the massive molecular data to phenotypic changes, in particular, at a network level, i.e., a novel method for identifying network biomarkers is in pressing need to accurately classify and diagnose diseases from molecular data and shed light on the mechanisms of disease pathogenesis. Rather than seeking differential genes at an individual-molecule level, here we propose a novel method for identifying network biomarkers based on protein-protein interaction affinity (PPIA), which identify the differential interactions at a network level. Specifically, we firstly define PPIAs by estimating the concentrations of protein complexes based on the law of mass action upon gene expression data. Then we select a small and non-redundant group of protein-protein interactions and single proteins according to the PPIAs, that maximizes the discerning ability of cases from controls. This method is mathematically formulated as a linear programming, which can be efficiently solved and guarantees a globally optimal solution. Extensive results on experimental data in breast cancer demonstrate the effectiveness and efficiency of the proposed method for identifying network biomarkers, which not only can accurately distinguish the phenotypes but also provides significant biological insights at a network or pathway level. In addition, our method provides a new way to integrate static protein-protein interaction information with dynamical gene expression data.

A novel mixed integer programming for multi-biomarker panel identification by distinguishing malignant from benign colorectal tumors

Multi-biomarker panels can capture the nonlinear synergy among biomarkers and they are important to aid in the early diagnosis and ultimately battle complex diseases. However, identification of these multi-biomarker panels from case and control data is challenging. For example, the exhaustive search method is computationally infeasible when the data dimension is high. Here, we propose a novel method, MILP_k, to identify serum-based multi-biomarker panel to distinguish colorectal cancers (CRC) from benign colorectal tumors. Specifically, the multi-biomarker panel detection problem is modeled by a mixed integer programming to maximize the classification accuracy. Then we measured the serum profiling data for 101 CRC patients and 95 benign patients. The 61 biomarkers were analyzed individually and further their combinations by our method. We discovered 4 biomarkers as the optimal small multi-biomarker panel, including known CRC biomarkers CEA and IL-10 as well as novel biomarkers IMA and NSE. This multi-biomarker panel obtains leave-one-out cross-validation (LOOCV) accuracy to 0.7857 by nearest centroid classifier. An independent test of this panel by support vector machine (SVM) with threefold cross validation gets an AUC 0.8438. This greatly improves the predictive accuracy by 20% over the single best biomarker. Further extension of this 4-biomarker panel to a larger 13-biomarker panel improves the LOOCV to 0.8673 with independent AUC 0.8437. Comparison with the exhaustive search method shows that our method dramatically reduces the searching time by 1000-fold. Experiments on the early cancer stage samples reveal two panel of biomarkers and show promising accuracy. The proposed method allows us to select the subset of biomarkers with best accuracy to distinguish case and control samples given the number of selected biomarkers. Both receiver operating characteristic curve and precision-recall curve show our method's consistent performance gain in accuracy. Our method also shows its advantage in capturing synergy among selected biomarkers. The multi-biomarker panel far outperforms the simple combination of best single features. Close investigation of the multi-biomarker panel illustrates that our method possesses the ability to remove redundancy and reveals complementary biomarker combinations. In addition, our method is efficient and can select multi-biomarker panel with more than 5 biomarkers, for which the exhaustive methods fail. In conclusion, we propose a promising model to improve the clinical data interpretability and to serve as a useful tool for other complex disease studies. Our small multi-biomarker panel, CEA, IL-10, IMA, and NSE, may provide insights on the disease status of colorectal diseases. The implementation of our method in MATLAB is available via the website: http://doc.aporc.org/wiki/MILP_k.

DISIS: prediction of drug response through an iterative sure independence screening

Prediction of drug response based on genomic alterations is an important task in the research of personalized medicine. Current elastic net model utilized a sure independence screening to select relevant genomic features with drug response, but it may neglect the combination effect of some marginally weak features. In this work, we applied an iterative sure independence screening scheme to select drug response relevant features from the Cancer Cell Line Encyclopedia (CCLE) dataset. For each drug in CCLE, we selected up to 40 features including gene expressions, mutation and copy number alterations of cancer-related genes, and some of them are significantly strong features but showing weak marginal correlation with drug response vector. Lasso regression based on the selected features showed that our prediction accuracies are higher than those by elastic net regression for most drugs.

Machine learning applications in cancer prognosis and prediction

Cancer has been characterized as a heterogeneous disease consisting of many different subtypes. The early diagnosis and prognosis of a cancer type have become a necessity in cancer research, as it can facilitate the subsequent clinical management of patients. The importance of classifying cancer patients into high or low risk groups has led many research teams, from the biomedical and the bioinformatics field, to study the application of machine learning (ML) methods. Therefore, these techniques have been utilized as an aim to model the progression and treatment of cancerous conditions. In addition, the ability of ML tools to detect key features from complex datasets reveals their importance. A variety of these techniques, including Artificial Neural Networks (ANNs), Bayesian Networks (BNs), Support Vector Machines (SVMs) and Decision Trees (DTs) have been widely applied in cancer research for the development of predictive models, resulting in effective and accurate decision making. Even though it is evident that the use of ML methods can improve our understanding of cancer progression, an appropriate level of validation is needed in order for these methods to be considered in the everyday clinical practice. In this work, we present a review of recent ML approaches employed in the modeling of cancer progression. The predictive models discussed here are based on various supervised ML techniques as well as on different input features and data samples. Given the growing trend on the application of ML methods in cancer research, we present here the most recent publications that employ these techniques as an aim to model cancer risk or patient outcomes.

Gene Selection Using Locality Sensitive Laplacian Score

Gene selection based on microarray data, is highly important for classifying tumors accurately. Existing gene selection schemes are mainly based on ranking statistics. From manifold learning standpoint, local geometrical structure is more essential to characterize features compared with global information. In this study, we propose a supervised gene selection method called locality sensitive Laplacian score (LSLS), which incorporates discriminative information into local geometrical structure, by minimizing local within-class information and maximizing local between-class information simultaneously. In addition, variance information is considered in our algorithm framework. Eventually, to find more superior gene subsets, which is significant for biomarker discovery, a two-stage feature selection method that combines the LSLS and wrapper method (sequential forward selection or sequential backward selection) is presented. Experimental results of six publicly available gene expression profile data sets demonstrate the effectiveness of the proposed approach compared with a number of state-of-the-art gene selection methods.

Edge biomarkers for classification and prediction of phenotypes

In general, a disease manifests not from malfunction of individual molecules but from failure of the relevant system or network, which can be considered as a set of interactions or edges among molecules. Thus, instead of individual molecules, networks or edges are stable forms to reliably characterize complex diseases. This paper reviews both traditional node biomarkers and edge biomarkers, which have been newly proposed. These biomarkers are classified in terms of their contained information. In particular, we show that edge and network biomarkers provide novel ways of stably and reliably diagnosing the disease state of a sample. First, we categorize the biomarkers based on the information used in the learning and prediction steps. We then briefly introduce conventional node biomarkers, or molecular biomarkers without network information, and their computational approaches. The main focus of this paper is edge and network biomarkers, which exploit network information to improve the accuracy of diagnosis and prognosis. Moreover, by extracting both network and dynamic information from the data, we can develop dynamical network and edge biomarkers. These biomarkers not only diagnose the immediate pre-disease state but also detect the critical molecules or networks by which the biological system progresses from the healthy to the disease state. The identified critical molecules can be used as drug targets, and the critical state indicates the critical point of disease control. The paper also discusses representative biomarker-based methods.

Pathway and network analysis in proteomics

Proteomics is inherently a systems science that studies not only measured protein and their expressions in a cell, but also the interplay of proteins, protein complexes, signaling pathways, and network modules. There is a rapid accumulation of Proteomics data in recent years. However, Proteomics data are highly variable, with results sensitive to data preparation methods, sample condition, instrument types, and analytical methods. To address the challenge in Proteomics data analysis, we review current tools being developed to incorporate biological function and network topological information. We categorize these tools into four types: tools with basic functional information and little topological features (e.g., GO category analysis), tools with rich functional information and little topological features (e.g., GSEA), tools with basic functional information and rich topological features (e.g., Cytoscape), and tools with rich functional information and rich topological features (e.g., PathwayExpress). We first review the potential application of these tools to Proteomics; then we review tools that can achieve automated learning of pathway modules and features, and tools that help perform integrated network visual analytics.

Network biomarkers reveal dysfunctional gene regulations during disease progression

Extensive studies have been conducted on gene biomarkers by exploring the increasingly accumulated gene expression and sequence data generated from high-throughput technology. Here, we briefly report on the state-of-the-art research and application of biomarkers from single genes (i.e. gene biomarkers) to gene sets (i.e. group or set biomarkers), gene networks (i.e. network biomarkers) and dynamical gene networks (i.e. dynamical network biomarkers). In particular, differential and dynamical network biomarkers are used as representative examples to demonstrate their effectiveness in both detecting early signals for complex diseases and revealing essential mechanisms on disease initiation and progression at a network level.

iPcc: a novel feature extraction method for accurate disease class discovery and prediction

Gene expression profiling has gradually become a routine procedure for disease diagnosis and classification. In the past decade, many computational methods have been proposed, resulting in great improvements on various levels, including feature selection and algorithms for classification and clustering. In this study, we present iPcc, a novel method from the feature extraction perspective to further propel gene expression profiling technologies from bench to bedside. We define ‘correlation feature space’ for samples based on the gene expression profiles by iterative employment of Pearson’s correlation coefficient. Numerical experiments on both simulated and real gene expression data sets demonstrate that iPcc can greatly highlight the latent patterns underlying noisy gene expression data and thus greatly improve the robustness and accuracy of the algorithms currently available for disease diagnosis and classification based on gene expression profiles.