Gene Expression Analysis Of Liver Cancer Assignment Sample

Introduction to Differential Gene Expression Analysis Of Liver Cancer Assignment Sample

Liver cancer affects many people, making it the sixth most common cancer and the third leading cancer death worldwide. In the past two decades, the number of new UK cases of liver cancer has doubled because of rising risks including hepatitis infection, issues related to drinking and non-alcoholic fatty liver disease. Even with better imaging and surgery, the chemical causes behind liver cancer are unclear. Exploring tumours’ transcripts with molecular profiling helps researchers spot genes that do not work properly which may be linked to cancer-causing, growth or resistance to treatment. It is possible to discover genes whose activity is strongly different in cancerous tissue and in normal tissue using differential gene expression analysis. Such types of genes can indicate the presence of disease early on and could be points for improving treatment with precision medicine.

The Cancer Genome Atlas (TCGA) supplies high-quality and comprehensive genetics information for several types of cancer, among them liver hepatocellular carcinoma (LIHC). Still, most studies that rely on TCGA data have not concentrated on the systematic comparison of gene expression between the normal and tumour tissue of liver cancer. The article uses strong bioinformatics techniques to analyse TCGA-LIHC data and aims to identify DEGs while examining their roles in liver cancer biology (Gao et al., 2021). The article describes a thorough analysis of differential gene expression based on the TCGA liver cancer dataset with the emphasis on the comparison of gene expression profiles in tumor and normal tissues. In the research, the top 20 genes showing the greatest variance among the samples are highlighted and these are assumed to hold the most promise in differentiating between cancerous and healthy tissues. Various statistical and graphic methods are used to investigate these genes, which include boxplot of expression values, sample clustering using PCA, density distribution plots, correlation heatmap, and comparison of mean expression.

Moreover, a random forest classification model is applied to analyze the predictive power of these top variable genes in calculated tumor and normal samples. The accuracy, misclassification rate and feature scores (importance) of the model constructed serve as additional confirmation of the discriminative ability of the selected genes. By covering these integrative efforts, the article will bring new knowledge about molecular signatures of liver cancer and emphasize the value of transcriptomic data in the discovery of cancer biomarkers (Zhu, and Zhang, 2025). The results can be used in the current endeavours to come up with efficient diagnostic measures and individualized medicine among patients with liver cancer.

This Differential Gene Expression Analysis of Liver Cancer Assignment Sample explores TCGA-LIHC data to identify genes that show different expression levels in cancerous and healthy liver tissues using bioinformatics techniques and statistical analysis. An Assignment writing helper can support students in understanding gene expression patterns, random forest classification, biomarker discovery, and the role of molecular profiling in cancer diagnosis and personalised medicine.

Materials and Methods

Source of Data and Summaries

The mRNA expression data analysed in this study was downloaded via The Cancer Genome Atlas (TCGA) Liver Hepatocellular Carcinoma (LIHC) cohort. This data has RNA sequencing expression profiles of liver cancer patients with tumor and adjacent normal tissues samples (National Cancer Institute, 2025). In particular, this study involved analysis of 424 samples comprising 374 tumor samples and 50 normal samples. The data was brought in a tabular .csv format with rows denoting genes and columns denoting individual samples.

Pre-Processing, and Sample Classification

The first step of data cleaning was to make the names of the genes the row identifiers and to make all the names of the genes unique with the help of the make.unique()function in R. Then, the column with identifiers of the genes was excluded to keep only the numerical values of the expression. Analysis of the data followed by transposing the dataset where necessary was done. The classification of the samples into the following groups: “Tumor” and “Normal” was made according to the TCGA barcode naming conventions. As per TCGA nomenclature, sample IDs with the prefixes “-01A-” are tumor tissues and the ones with prefixes “-11A-” are normal tissues (Wilson et al., 2017). This made it possible to give each sample a status label which was then used to create a categorical variable that had two levels; Normal (n=50) and Tumor (n=374).

Highly Variable Genes Selection

Variance among samples was determined in order to concentrate the analysis on genes that would be the most informative. This step played a key role in the determination of genes which had a significant variation in their expression in the entire dataset. To assume that the more variable the expression of a gene, the more promising a biological and clinical relevant gene, the top 20 most variable genes were used in downstream analysis.

Exploratory Data Visualization

Several statistical plots were created with ggplot2 to get a visual impression of the differences in the patterns of gene expression between the tumor and normal samples.

Box Plots

The top 20 variable genes expression was reshaped into a long format with pivot_longer() of tidyverse package and became appropriate to be displayed in a grouped fashion. Boxplots were then created having gene names as x-axis and expression values as y-axis and different color fill according to group (Tumor vs Normal). This enabled comparison of the distribution of gene expression in the two groups directly.

Density Plots

Density plots of the top 20 genes were created to investigate the distribution of the values of the expression in general. They employed smoothed histograms on these plots and employed them to compare the shape and spread of the expression levels in the two groups. The density plots gave an overview of expression differences and possible skew or variability in the data across the world.

Comparison of Mean Expression

In order to further measure group-wise differences of expression, the expression value of each gene was averaged individually in Tumor and Normal samples. Bar graphs were generated next to depict the average expressions among groups. This gave a reduced and quantitative perspective of variation in expression throughout the chosen genes.

Correlation Heatmap

The heatmap package was used to create a heatmap that would be used to interpret the pairwise correlation of the top 20 variable genes among all the samples. A correlation matrix was used to produce the heatmap which also used hierarchical clustering to identify patterns of co-expression among the genes. The heatmap applied the distance measure of correlation on both the rows and columns, and this measure focuses on the relative similarities of the gene expression profiles.

Principal Component Analysis (PCA)

The scaled and transposed dataset was subjected to Principal Component Analysis (PCA) to fewer dimensions and cluster visualization patterns. The plot of the first two principal components (PC1 and PC2) was generated with ggplot2 and the color coding according to the sample status (Tumor or Normal). The analysis was done so that the natural group separation could be visualized and to assist in determining whether the patterns of gene expression would be able to separate the two classes.

Machine Learning Classifier: Random Forest

In order to determine the top 20 variable genes' predictive ability to distinguish between samples classified as tumor or normal, a Random Forest classification model was fitted via the randomForest package. The model fit relied on the best genes as predictor variables and sample status as the response variable. The forest grew 500 trees, and at each split four variables were tested. The Out-of-Bag (OOB) error rate and confusion matrix were used to evaluate the work of the model. The important measures that came out of the model were class-wise error rates, and variable importance scores (Bartha, and Győrffy, 2021). The importance of variables was plotted with the help of varImpPlot() function in order to observe the genes that had the most significant contribution to the classification problem.

Software and Environment

All the R code was written in R (version 4.2 or higher) on RStudio environment and handled all the preprocessing of data, statistical analysis, visualization, and machine learning. The R packages used were as following:

• tidyverse: to wrapper data manipulation and reshaping
• ggplot2: to plot graphics
• heatmap: to plot heatmap
• randomForest: in classification modeling

Unless stated otherwise default settings were employed. The whole analysis was repeatable and done locally on a machine with sufficient hardware to analyze bulky genomic data.

Ethical considerations

The data analyzed in the presented study was obtained through the widely available TCGA repository, which adheres to open-access guidelines. This secondary data analysis did not need any ethical approval because the data used does not involve any personally identifiable information.

Results

This section describes the overall results of the differential gene expression analysis performed on TCGA liver cancer dataset. A comparative study between tumor and normal liver tissues was conducted using a combination of statistical, visual, and machine learning techniques. The 20 most variably expressed genes were analyzed mainly to determine the suitability of these genes to serve as discriminative biomarkers of hepatocellular carcinoma.

Samples Classification and Grouping Distribution.

The variance was computed across all genes and the entire dataset in order to limit the number of genes to those with the most informative expression pattern. The genes with greater variance were deemed to have greater probability of developing differential expression between tumor and normal samples and therefore presented the best target genes in the process of biomarker discovery. To perform deep analysis, the 20 most variably expressed genes were picked. These genes may be taken in the most biologically interesting variations in transcriptomic action relating to liver cancer (DataCamp, 2025). This criterion of variance in selection of genes made analysis to focus on genes with changing expression patterns and not background noise or housekeeping genes that are uniformly expressed.

Patterns of Expression through Box Plot Analysis

Expression distributions of the 20 most changeful genes were visualized as boxplots and provided a good overview of medial expressions, interquartile ranges, and voltage outliers of each individual gene in both the tumor and convening samples. The boxplot plots showed meaningful differences between the two groups regarding the level of expression. Median values of tumor samples were high than those of the convening samples for a great reckon of genes, indicating an overexpression design of genes in crab tissues. Moreover,' there are aggregated genes that had a wider interquartile range in tumor samples, which is suggestive of more changeful expression. Such an observance may be explained by genetics and constitution heterogeneousness store in the tumor microenvironment (GeeksforGeeks, 2020). In addition as well as the outliers observed among tumor samples in some of the genes suggested boringness of subpopulations inside the tumor group which could have clear cut building profiles, and which could be related to accusative characteristics of the disease like disease stage or reaction to treatment.

Density Plot

A Density plot was also produced on the building data of the same 20 genes to add to the findings of the boxplot. This was aimed at evaluating the universal form and width of gene building distributions in tumor and convening samples in a smoothed form. The density curves showed that tumor samples were shifted to the right implying that the overall gene expression levels in cancer tissues were greater. In sharp contrast, normal samples had a symmetrical and narrower distribution (Holtz, 2025). The fact that the two curves did not overlap much implied that the genes chosen did indeed distinguish between the two types of tissues in a consistent manner. This finding was quite consistent with the expectation that tumor tissues have a distorted transcriptomic landscape, which in most cases may present as an upregulated oncogene or downregulated tumor suppressor gene. The density plot therefore supported the earlier conclusions made by the boxplot analysis, in that, there were significant expression

Quantitative Variations in Comparison of Mean Expression

In order to numerically describe the patterns of expression visualized in the previous plots, the average expression value of each of the 20 most popular genes was calculated in tumor and normal samples separately. The visualization of these means was then performed in bar charts to enable direct comparison in numerical terms. Bar plots were used to simply explain the average gene activity, and it was clear that most of the genes were expressed at higher means in tumor tissues (Murraylax, 2015). The average expression in tumor samples was higher than the one in normal samples in certain cases by more than twice, which supports the idea that these genes could be upregulated in order to respond to oncogenic stimuli. That these differences were not singular artifacts was suggested by the consistency of the heightened expression in the various formats of visualization, suggesting that they represented a real biological difference. This numerical analysis provided additional evidence to the argument that these genes can become effective markers in the diagnostics and classification of liver cancer.

Correlation Mapping of Co-expression of the Genes

A correlation matrix was calculated and represented in a heatmap to explore possible co-expression relationships among the top variable genes. This analysis gave an idea about the simultaneous expression of particular genes, which may indicate their common regulation or involvement into the same signaling pathways. The heatmap showed several cases of high positive correlation and some pairs of genes were even correlated with coefficients close to 0.9. Such high correlation values reflected genes that could be co-regulated by transcription factors of similar type or could be working in the same biological process like cell proliferation, angiogenesis or immune response. The heatmap hierarchical clustering clustered together genes that had similar profiles in terms of their correlation pattern, and it was possible to identify gene modules that may be functionally related units (Sthda, 2025). Such results were especially informative in regards to the general transcriptomic structure of liver cancer because they indicated that the phenomenon of differential expression might not exist in the vacuum but rather as a coordinated network response.

Principal Component Analysis (PCA)

The high-dimensional gene expression data was reduced to a lower-dimensional space with Principal Component Analysis (PCA) being done, thereby keeping as much variance as possible. The two initial principal components were used to plot the visualization of the goodness of clustering of the samples on the basis of their expression profiles. The resultant PCA plot identified a clear distinction between the tumor and normal samples (Szczesna, 2022). Most of the tumor samples clustered together representing a concentrated cluster, whereas the normal samples were in another area of the PCA space. There were a few tumor samples that overlapped with the normal sample cluster and this may be explained by the heterogeneity of tumor composition or by possible misclassification. PC1 accounted for the greatest percentage of variance, implying that it was the most informative dimension that differentiated between the two groups. The resolution of this division showed that even a small number of highly variable genes could efficiently bottle-neck the underlying biological distinctions between tumor and non-tumor liver tissues.

Predictive modelling - Random forest Classifier

In order to evaluate how well the top 20 genes can be used as a classification attribute, a Random Forest (RF) model was trained with these 20 genes as input features and sample status (tumor or normal) as an output class. It was made up of 500 decision trees and four features were looked at during each split. The model was determined to have an Out-of-Bag (OOB) error rate of 4.95% which implied a high level of accuracy in sampling classification. As interpreted by the confusion matrix, this model had no problem classifying 372 tumor samples accurately and the class error was 0.53%. It however missed classification in 19/50 of the normal samples giving a class error of 38% on the normal group. The difference in classification accuracy between both groups might be partly due to the class imbalance which tends to make the model favour the majority class (Tutorialspoint, 2025). Nevertheless, the model performance overall was high, particularly on tumor samples recognition, which represent the main clinical interest in cancer detection.

In order to check which genes had the most significant contribution to the classification task, a variable importance plot was created using the Random Forest model. This plot performed a ranking of the genes based on Mean Decrease in Gini Index which is a measure that showed the extent to which a gene enhanced the purity of the splits in the decision tree. What the plot demonstrated is that there was a group of genes with significantly higher importance scores than others, meaning that they were at the center of determining the class of a sample (GeeksforGeeks, 2020). These highly ranking genes matched very well with the genes that had depicted the most difference in expression and shown the highest correlation values in the previous analyses. The combination of evidence by use of several analytical techniques gave significant reasons as to why these genes should be at the forefront in the remainder of the story. Their strong showing in statistical, visual and machine learning analyses rendered them promising as either diagnostic markers or therapy targets in liver cancer.

Synthesized Discussion of Findings

Overall, the findings of this paper draw a logical and convincing image of liver cancer differential gene expression. The 20 most variable genes were both highly differing in their expression between tumor and normal tissues as well as exhibiting high internal consistency and predictive value. The different plots, box plots, density plots, mean expression charts, heatmaps and PCA diagrams spoke the same story of high and varying expression in tumor samples. Random Forest model provided an extra level of functional validation as it showed that these genes were able to classify samples, and especially the tumor group with high accuracy (Gao et al., 2021). Even more, identification of highly correlated modules of gene regulation implied that liver cancer might not only be accompanied by individual alterations of gene expression, but by coordinated perturbations of regulatory networks.

The analysis presented a complex picture of the liver cancer biology, which highlights the potential of transcriptomic data in the discovery of important molecular aspects of disease. The information obtained here provides a basis of more focused research on the functions of these genes alone and in combination in the development and growth of the cancer.

Discussion

Results of the present analysis represent valuable information about molecular differences between tumor and normal liver tissues according to the gene expression profiles based on the Liver Hepatocellular Carcinoma dataset of TCGA. By targeting the 20 most variable genes among samples, the study could also show that a comparatively small number of genes can exhibit strong discriminatory value and biological concern when it comes to the case of hepatocellular carcinoma (HCC). The major interpretations, biological importance, methodological advantages and drawbacks, and the possible implications of the results concerning the further research and clinical practice are discussed and elaborated in this discussion.

Differential Expression Patterns interpretation

The up-regulation nature of the top 20 genes in tumor tissues based on the boxplots, density plot, and mean expression value analysis is highly indicative of the possibility of these genes being upregulated due to the oncogenic transformations. Widespread transcriptional reprogramming of tumor cells helps to promote rapid proliferation, apoptosis evasion, maintenance of angiogenesis and immune system modulation. Such malignant adaptations are reflected in the pattern of expression in this study. A fraction of the best variable genes will represent known oncogenes or transcription factors, or cell cycle regulation genes (Wu et al., 2021). Their upregulation in tumor specimens may be a sign of their active participation in tumor development. Others could be new or less characterized genes whose pattern of expression had not been linked to liver cancer before. This twofold possibility to replicate known drivers and identify new candidate genes makes the analysis rich and suggests additional possibilities of further investigations.

Besides, the significantly different interquartile ranges and the outliers in the tumor samples indicate the heterogeneity of liver cancer. Tumors are not clonal entities but they are composed of several subclonal populations that have diverse genetic and phenotypic features. The inter-sample variation in gene expression in tumor samples is probably a reflection of the variation in tumor stage, state of differentiation, the underlying etiological factors like viral infection versus alcohol intake and individual patient immune response (Cao et al., 2021). The causes of such variability must be understood to progress to personalized medicine where, depending on the molecular profile of an individual tumor, different treatments can be used.

Correlation and Co-expression Structures

The correlation heatmap helped to introduce an extra dimension of knowledge regarding the interaction of the top 20 genes in the context of the entire transcriptome. Positive correlations found between some genes could indicate that they are co-regulated (perhaps by common upstream transcription factors, or because they lie in the same signalling pathway, or because they are co-localised on the chromosome). Discovery of these relationships is important in explaining the gene regulatory networks which are perturbed in cancer. Cluster of genes that are highly correlated could also indicate biological modules. Imagine that a number of genes, which have strong inter-correlations, are known to be engaged in mitotic spindle assembly or DNA repair; then this may indicate that a particular biological process is under up regulation in tumor cells (Alves et al., 2021). This kind of information can be used to direct specific studies aimed at confirming that these pathways are core to the occurrence and progression of liver cancer. Besides, correlation structures can be used to develop biomarker panels. This is in place of using one marker; diagnosis specificity and sensitivity could be raised by using genes that act in a tightly regulated module.

Different Clustering Tends Unveiled by PCA

Principal Component Analysis (PCA) gave strength to the observation that tumor and normal liver tissues reside in different locations in the gene expression space. The strong distinction evident within the initial two main components indicates that the largest or most significant sources of variance within the information are motivated by biology distinction amid the tumor and normal conditions. This finding corresponds with the findings of other cancer investigators, in which PCA has shown to be useful in separating malignant and non-malignant samples based on expression data.

The clustering pattern in PCA also confirms the construct that the top 20 genes identified are good enough to represent important aspects of the tumorigenic process. Although PCA is an unsupervised method and does not use the class labels, the fact that it can separate the samples in such a clear way means that the natural structure of the data coincides with clinical classes (Wang et al., 2025). This observation increases the belief in the biological significance of the chosen genes and the reasonability of their further consideration as possible biomarkers. It also bears notice that some of the tumor samples were found to be near the normal sample cluster in PCA space. This may be attributed to a number of reasons. First, these samples can be low-stage tumors whose gene expression patterns have not yet become significantly different to those of normal tissues (Hong et al., 2025). Second, certain samples can include large fractions of non-tumor cells (e.g. stromal or immune cells) which will result in a more normalised expression signature. Third, clustering patterns can be affected by technical variability or batch effects that can be reduced by preprocessing but in some cases cannot be ignored. These borderline cases could be worth investigating to gain valuable knowledge about tumor evolution and the process of a premalignant to a malignant state.

Machine Learning Predictive Utility

Random Forest classification model had high accuracy in differentiation of tumor and normal samples especially in the identification of tumor cases. The small misclassification rate of tumor samples (23.3%) indicates that the top 20 genes are strong predictors of malignancy. This is especially relevant in clinical practices, where false negative results, which identify a tumor as being normal, pose great dangers to patient outcomes (Tang et al., 2024). Higher error rate on normal samples in the model is significant, but not as serious in the setting where the task is to detect cancer, but it indicates that class balancing or different algorithms should be considered in future research.

Random Forest will be applicable in this task since it is apt to deal with nonlinear relationships as well as interactions between features without prior assumptions. It is also not susceptible to overfitting due to use of multiple decision trees especially when the samples are few compared to the predictor variables. The variable importance analysis also provided interpretability to the model, that is, which genes had the greatest influence on the classification process. The genes that have high scores of Mean Decrease in Gini are not only useful in prediction, but also would likely have biologically significant roles in cancer development (Bao, and Wu, 2024). Although the performance of Random Forest was satisfactory, in further research, it may be useful to compare the performance of this model with other classification models, including Support Vector Machines (SVM), logistic regression, or deep learning methods. All of these procedures have their strong points and ensemble methods may perhaps achieve higher accuracies by pooling predictions of many algorithms.

Liver Cancer Biology

Liver executes numerous physiological activities and any disturbance in its cellular milieu can promptly result in pathological effects. Hepatocellular carcinoma usually develops on the backdrop of chronic liver disease, such as cirrhosis and viral hepatitis. The changes in gene expression observed in this manuscript are possibly a consequence of not only the tumorigenic process but also the underlying inflammatory and fibrotic landscape (Caputo et al., 2023). A number of the discovered genes can be related to the process of angiogenesis, immune evasion, or metabolic reprogramming. Although they were not performed in the given study, functional annotation and pathway enrichment analyses may allow placing these genes into the context of particular biological processes. This type of analyses would give a mechanistic component to the results and could potentially identify novel therapeutic targets.

The overexpression of some of the genes that is observed in the tumors may be correlated with the drug resistance or sensitivity. As an example genes pertaining to DNA repair processes can be associated with chemotherapy resistance whereas genes impacting cell cycle checkpoints can be relevant to targeted therapy response (Wang et al., 2021). Incorporation of the gene expression data with clinical outcome data including survival rates, response to treatment and recurrence would go a long way in making this work translationally important.

Limitation and Considerations

Even though the findings are quite solid, one should take into consideration several limitations. Firstly, the research utilised only the secondary data provided by the TCGA which is of high quality but could be potentially affected by batch effects, technical variation, and sampling biases. Another issue is the imbalance between the tumor and normal samples, which may affect the model, both statistical and machine learning performance (Zhang et al., 2025). Such methods as class weighting and synthetic oversampling were not applied in this research, but they might enhance model fairness in future research.

The lack of clinical annotation, e.g. tumor grade, patient age, gender or etiology is another limitation. Such metadata inclusions might allow more fine-grained analyses and the identification of gene expression signatures particular to some subgroup. In addition, the analysis was limited to the examination of the expression levels and did not apply to other aspects of DNA methylation, copy number variation, and mutation status. A combination of multi-omics data would provide a more comprehensive picture of the molecular landscape and would help reveal interactions between various layers of regulation (Yi et al., 2022). Lastly, this computational study was not able to functional validate the identified genes. Although statistical and visual evidence is quite convincing about the relevance of these genes, experimental work needs to be done to prove biological roles of these genes and their therapeutic applicability. Next techniques to be applied in the validation of the expression and role of these genes in liver cancer models would include qPCR, CRISPR-Cas9 gene editing, and immunohistochemistry techniques.

Discussion and Prospects

The practical consequences of this work are great both in terms of fundamental research and clinical practice. In the research view, identification of highly variable and differentially expressed genes has a basis of generating hypotheses. The functional studies can be further carried out on these genes to identify their functions in liver cancer development and sustenance (Yan et al., 2025). Identification of their mechanistic functions may point out novel therapeutic targets and enhance disease comprehensions.

The capacity to differentiate tumor and normal tissue with a small gene panel in the clinical setting allows the possibility of creating diagnostic assays. This is because such assays may be applied in the early detection, risk stratification, or response to treatment monitoring. Moreover, adding the expression profiles to precision medicine models may aid in pairing the patients with the treatments that target the molecular makeup of their tumor (Fend et al., 2022). In future, it would be useful to confirm the results in independent cohorts or in other publicly available datasets e.g. International Cancer Genome Consortium (ICGC) or Gene Expression Omnibus (GEO). The generalizability of the results would be reinforced by replication in dissimilar populations. Also, it may be interesting to combine the transcriptomic data with clinical outcome data to develop prognostic models allowing to predict the risk of survival or recurrence on the basis of gene expression signature.

Conclusion

This paper identifies the significant differences in the gene expression between the tumor and normal liver tissues based on TCGA data, and the analysis is done on 20 most variable genes. These genes were found to have high discriminatory ability using statistical analysis, visualization and machine learning, which holds promise as diagnostic biomarkers. The data confirm the informative value of transcriptomic profiling in the study of hepatocellular carcinoma and highlight the integrative value in the molecular target identification. Although additional validation is necessary, the study forms a solid background to further research in the field of early detection, classification, and individual treatment approaches in liver cancer.

Read Our Last Sample: https://www.newassignmenthelp.co.uk/principles-of-health-and-social-care-essay-sample-52383

Reference List

• Alves AF, Moura AC, Andreolla HF, et al. Gene expression evaluation of antioxidant enzymes in patients with hepatocellular carcinoma: RT-qPCR and bioinformatic analyses. Genet Mol Biol. 2021;44(2):e20190373. Published 2021 Apr 2. https://doi.org/10.1590/1678-4685-GMB-2019-0373
• Bao M, Wu A. Understanding the heterogeneity in liver hepatocellular carcinoma with a special focus on malignant cell through single-cell analysis. Discov Oncol. 2024;15(1):241. Published 2024 Jun 24. https://doi.org/10.1007/s12672-024-01115-9
• Bartha Á, Győrffy B. TNMplot.com: A Web Tool for the Comparison of Gene Expression in Normal, Tumor and Metastatic Tissues. Int J Mol Sci. 2021;22(5):2622. Published 2021 Mar 5. https://doi.org/10.3390/ijms22052622
• Cao J, Zhang C, Jiang GQ, et al. Identification of hepatocellular carcinoma-related genes associated with macrophage differentiation based on bioinformatics analyses. Bioengineered. 2021;12(1):296-309. https://doi.org/10.1080/21655979.2020.1868119
• Caputo WL, de Souza MC, Basso CR, Pedrosa VA, Seiva FRF. Comprehensive Profiling and Therapeutic Insights into Differentially Expressed Genes in Hepatocellular Carcinoma. Cancers (Basel). 2023;15(23):5653. Published 2023 Nov 30. https://doi.org/10.3390/cancers15235653
• Descriptive Statistics in R. DataCamp. Published 2025. Accessed June 19, 2025. https://www.datacamp.com/doc/r/descriptives
• Feng ZH, Liang YP, Cen JJ, et al. m6A-immune-related lncRNA prognostic signature for predicting immune landscape and prognosis of bladder cancer. J Transl Med. 2022;20(1):492. Published 2022 Oct 29. https://doi.org/10.1186/s12967-022-03711-1
• Gao S, Gang J, Yu M, Xin G, Tan H. Computational analysis for identification of early diagnostic biomarkers and prognostic biomarkers of liver cancer based on GEO and TCGA databases and studies on pathways and biological functions affecting the survival time of liver cancer. BMC Cancer. 2021;21(1):791. Published 2021 Jul 8. https://doi.org/10.1186/s12885-021-08520-1
• Gao X, Zhao C, Zhang N, et al. Genetic expression and mutational profile analysis in different pathologic stages of hepatocellular carcinoma patients. BMC Cancer. 2021;21(1):786. Published 2021 Jul 8. https://doi.org/10.1186/s12885-021-08442-y
• GeeksforGeeks. Boxplots in R Language. GeeksforGeeks. Published April 12, 2020. Accessed June 14, 2025. https://www.geeksforgeeks.org/boxplots-in-r-language/
• GeeksforGeeks. Random Forest Approach in R Programming. GeeksforGeeks. Published May 31, 2020. Accessed June 14, 2025. https://www.geeksforgeeks.org/random-forest-approach-in-r-programming/
• Holtz Y. Density chart with several groups. R-graph-gallery.com. Published 2025. Accessed June 14, 2025. https://r-graph-gallery.com/135-stacked-density-graph.html
• Hong Y, Wang D, Liu Z, Chen Y, Wang Y, Li J. Decoding per- and polyfluoroalkyl substances (PFAS) in hepatocellular carcinoma: a multi-omics and computational toxicology approach. J Transl Med. 2025;23(1):504. Published 2025 May 2. https://doi.org/10.1186/s12967-025-06517-z
• Murray JM. Bar Plots to Illustrate Means. Murraylax.org. Published 2015. Accessed June 14, 2025. https://murraylax.org/rtutorials/barplots.html
• National Cancer Institute. Genomic Data Commons Data Portal. Published 2025. Accessed June 14, 2025. https://portal.gdc.cancer.gov/
• ‌STHDA. ggplot2 : Quick correlation matrix heatmap - R software and data visualization - Easy Guides - Wiki - STHDA. Sthda. Published 2025. Accessed June 14, 2025. https://www.sthda.com/english/wiki/ggplot2-quick-correlation-matrix-heatmap-r-software-and-data-visualization
• Szczęsna K. Unsupervised Learning - Principal Component Analysis (PCA). Rpubs.com. Published February 5, 2022. Accessed June 14, 2025. https://rpubs.com/KarolinaSzczesna/862710
• Tang J, Peng X, Xiao D, Liu S, Tao Y, Shu L. Disulfidptosis-related signature predicts prognosis and characterizes the immune microenvironment in hepatocellular carcinoma. Cancer Cell Int. 2024;24(1):19. Published 2024 Jan 9. https://doi.org/10.1186/s12935-023-03188-y
• Tutorialspoint. R - Random Forest. Published 2025. Accessed June 14, 2025. https://www.tutorialspoint.com/r/r_random_forest.htm
• Wang H, Feng C, Lu M, et al. Integrative single-cell transcriptome analysis reveals a subpopulation of fibroblasts associated with favorable prognosis of liver cancer patients. Transl Oncol. 2021;14(1):100981. https://doi.org/10.1016/j.tranon.2020.100981
• Wang L, He X, Shen Y, et al. A Novel Ferroptosis-Related Gene Prognosis Signature and Identifying Atorvastatin as a Potential Therapeutic Agent for Hepatocellular Carcinoma. Curr Issues Mol Biol. 2025;47(3):201. Published 2025 Mar 18. https://doi.org/10.3390/cimb47030201
• Wilson S, Fitzsimons M, Ferguson M, et al. Developing Cancer Informatics Applications and Tools Using the NCI Genomic Data Commons API. Cancer Res. 2017;77(21):e15-e18. https://doi.org/10.1158/0008-5472.CAN-17-0598
• Wu R, Guo W, Qiu X, et al. Comprehensive analysis of spatial architecture in primary liver cancer. Sci Adv. 2021;7(51):eabg3750. https://doi.org/10.1126/sciadv.abg3750
• Yan W, Xiao GH, Wang LJ, Zhou Y, Yang F, Mou KH. CAFs activated by YAP1 upregulate cancer matrix stiffness to mediate hepatocellular carcinoma progression. J Transl Med. 2025;23(1):450. Published 2025 Apr 16. https://doi.org/10.1186/s12967-025-06325-5
• Yi Q, Peng J, Xu Z, et al. Spectrum of BRAF Aberrations and Its Potential Clinical Implications: Insights From Integrative Pan-Cancer Analysis. Front Bioeng Biotechnol. 2022;10:806851. Published 2022 Jul 14. https://doi.org/10.3389/fbioe.2022.806851
• Zhang Y, Zhang H, Liu L. Integration of single-cell and bulk RNA sequencing identifies and validates T cell-related prognostic model in hepatocellular carcinoma. PLoS One. 2025;20(5):e0322706. Published 2025 May 2. https://doi.org/10.1371/journal.pone.0322706
• Zhu Y, Zhang P. Gene expression profile of anoikis reveals new subtypes of liver cancer and discovery of therapeutic targets and biomarkers. Sci Rep. 2025;15(1):12740. Published 2025 Apr 13. https://doi.org/10.1038/s41598-025-96488-4