2.1 Genomics

Genomics constitutes a comprehensive examination of an organism's genome, involving the sequencing, mapping, and analysis of its DNA [35]. Advancements in technology related to DNA sequencing, particularly NGS, facilitate extensive analysis of entire genomes or specific regions, thereby enabling the identification of mutations such as single-nucleotide polymorphisms (SNPs), insertions, deletions, copy number variations (CNVs), and structural modifications that are crucial to the development of cancer [36]. Techniques including whole-genome sequencing (WGS), single-cell DNA sequencing, and targeted gene panels empower researchers to precisely identify mutations that can inform therapeutic strategies. For instance, mutations in genes such as KRAS, BRAF, and TP53 correlate with specific types of cancer and therapeutic responses, underscoring the clinical significance of genomic profiling [37, 38]. The intricate nature of genomic data requires sophisticated bioinformatics support for comprehensive analysis. Specialized computational platforms are indispensable for the interpretation of variants. Bioinformatics tools, such as GATK, are recognized as leading solutions for the detection of SNPS and small insertions and deletions, while MuTect specializes in identifying somatic mutations in paired tumor-normal samples [39, 40]. ANNOVAR enhances these methodologies by offering detailed variant annotations based on functional consequences [41]. Moreover, single-cell genomics tools, such as CellRanger, have revolutionized gene expression analysis at the individual cell level, essential for understanding cancer heterogeneity and evolution [42]. Tools like MORPHEUS and Monocle 3 facilitate interactive clustering of gene expression, lineage tracing, and trajectory analysis, all of which are vital for unraveling cancer progression and cellular differentiation [43, 44]. By integrating these bioinformatics instruments to convert raw genetic data from DNA sequencing techniques into meaningful clinical insights, researchers can uncover the prognostic and predictive value of specific genetic alterations, ultimately advancing personalized approaches in oncology.

2.2 Transcriptomics

Transcriptomics provides an ongoing perspective on gene expression in cells, offering vital insights into the molecular mechanisms underlying cancer development and progression [45]. Transcriptomics focuses on analyzing RNA transcripts, such as mRNA, noncoding RNA, and miRNA, to assess gene expression variances between cancerous and healthy tissues. Advanced technologies like RNA-seq and microarrays have reinvented gene expression profiling, revealing patterns associated with different tumor types, stages, and responses to treatment [46, 47]. For example, the high and low expression of certain oncogenes and tumor suppressor genes respectively can serve as diagnostic or prognostic biomarkers, enhancing our understanding and potentially aiding in mitigating disease progression [48, 49].

Bioinformatics tools play a vital role in analyzing transcriptomic data [50]. Differential gene expression analysis is key for identifying genes that are either upregulated or downregulated in cancer, supported by statistical model-based tools such as DESeq2 and EdgeR that work with count-based RNA-seq data [21, 51]. While limma was originally created for microarray research, it is now commonly used in RNA-seq to uncover significant expression patterns [52]. Multiomics integration platforms like Sangerbox 3, MOFA+, and MixOmics improve the visualization of relationships between datasets [53-55]. For single-cell RNA-seq analysis, tools such as Scanpy and Seurat aid in clustering and in-depth examination of expression variations, providing insights on the cellular heterogeneity of cancer [56-58]. Additionally, platforms like pyBioPortal and cBioPortal enable the integration of transcriptomic data with clinical and genomic information, enhancing the overall understanding of cancer biology [59, 60].

For RNA-seq quantification, Salmon and Kallisto utilize ultra-fast, alignment-free techniques to estimate precise transcript numbers, while STAR and HTSeq offer alignment and read counting for gene expression studies [61-64]. In addition, single-cell-specific tools like CellBender improve data quality by filtering out ambient RNA, and deep learning models such as scVI support imputation and clustering analyses [65, 66]. Furthermore, Dyngen and scRNASeqDB have advanced single-cell research methods which facilitate gene expression data simulation and provide dynamic platforms for cross-cancer exploration [67-69]. Tools like scissor link single-cell RNA-seq data to clinical outcomes, integrating cellular insights with survival rates and treatment responses, thus paving the way for translational oncology applications [70]. As technological advancements progress, transcriptomics leads the way in precision oncology, presenting exceptional opportunities to decode cancer's molecular complexities and create targeted therapeutic interventions.

2.3 Proteomics

Proteomics is the comprehensive study of all proteins generated by a genome and plays a crucial role in biomedical research, notably in precision oncology [71]. Unlike genomics and transcriptomics, which infer gene expression mainly from genetic sequences, proteomics seeks to reveal the functional consequences of genes: specifically, at the protein level that affect cellular operations and define phenotypes [72]. Additionally, proteomics facilitates the direct quantification of protein levels and critical posttranslational modifications (PTMs) such as phosphorylation, glycosylation, and ubiquitination, offering valuable insights into the functional status of proteins linked to cancerous changes. Proteins discovered through proteomic analyses can provide important information about tumor characteristics, patient outlook, and possible treatment responses. For example, well-known biomarkers such as CA-125 and PSA are used in ovarian and prostate cancers, respectively [73, 74]. By integrating proteomic assessments into clinical workflows, healthcare providers can modify treatment plans according to a patient's molecular profile, ultimately improving treatment effectiveness and patient outcomes.

Recent advancements in mass spectrometry (MS) and protein microarray technologies have positioned proteomics at the forefront of scientific inquiry [75]. Techniques such as liquid chromatography–tandem mass spectrometry (LC–MS/MS) are widely utilized for high-throughput protein analysis [76]. These sophisticated methods possess the capability to detect thousands of proteins within complex biological samples and quantify their abundance, thereby facilitating comprehensive tumor proteome profiling. Protein microarrays, an additional critical tool in the field of proteomics, allow for the simultaneous investigation of multiple proteins and their interactions with various molecules [77]. By presenting arrays of known proteins on a solid substrate, this technology enables high-throughput screening of protein expression, interactions, and functions in an exceptionally efficient and informative manner.

To effectively utilize proteomics in oncology, it is essential to have robust bioinformatics tools for managing and analyzing the vast data generated by these studies. As the field advances, computational platforms that support biomarker discovery have become increasingly important. Software like MaxQuant and Skyline enables precise quantification of protein levels and PTMs, allowing for comparative analyses between tumor and normal tissues or between sensitive and resistant tumors [78, 79]. Moreover, tools like Proteome Discoverer and PeptideShaker streamline workflows for protein identification, while OpenMS provides an open-source solution for LC–MS data analysis [80-83]. For exploring protein interactions, tools such as STRING, Cytoscape, and BioGRID aid in mapping complex protein networks, and PhosphoSitePlus offers a comprehensive catalog of verified PTMs [84-87]. Structural prediction software, including Alphafold and I-TASSER, reveals insights into protein function, supported by visualization tools like Pymol [88-90]. High-throughput tools such as MS-DIAL and DIA-NN, alongside ML applications like DeepNovo, improve peptide identification and sequencing precision [91-93]. Additionally, statistical analysis platforms like Perseus and limma/EdgeR in R allow for detailed interpretation of proteomic data, ensuring its biological relevance while minimizing false discoveries [94, 95].

The integration of these advanced tools facilitates a substantial enhancement in our comprehension of the molecular foundations of cancer, concurrently contributing to the development of molecular biomarkers, targeted therapies, and personalized treatment strategies.

2.4 Epigenomics

Cancer develops not solely from genetic mutations but also from significant epigenetic alterations that impact gene expression and cellular behaviors, including uncontrolled growth, invasion, and metastasis [96]. For instance, the hypermethylation of tumor suppressor genes may reduce their expression, whereas hypomethylation can potentially activate oncogenes, thereby contributing to malignancy. These epigenetic processes are essential for the identification of biomarkers that can inform personalized treatment strategies. Epigenomics, which investigates epigenetic modifications of genetic material, represents an emerging field in precision oncology [97]. In contrast to genomics, which focuses on DNA sequences, epigenomics examines how alterations such as DNA methylation, histone modifications, and noncoding RNA interactions influence gene expression. Importantly, epigenetic changes are often reversible, rendering them promising targets for therapeutic intervention. Currently, a variety of epigenetic modulators capable of altering these changes are undergoing evaluation in clinical trials, emphasizing the importance of epigenomics in cancer care [98].

Recent advances in high-throughput sequencing technologies have revolutionized epigenomic research [99]. Methods like whole-genome bisulfite sequencing (WGBS), chromatin immunoprecipitation sequencing (ChIP-seq), and RNA-seq are frequently used to map DNA methylation patterns, histone modifications, and interactions with noncoding RNAs [100-103]. For example, WGBS provides a comprehensive perspective on DNA methylation across the genome by distinguishing between methylated and unmethylated cytosines, thereby revealing methylation patterns associated with various cancers. This technique is crucial for understanding the epigenomic landscape of tumors and for identifying methylation signatures that may serve as biomarkers. ChIP-seq enables researchers to investigate histone modifications and transcription factor binding sites throughout the genome, offering vital insights into how these changes affect gene expression in cancer. This approach can uncover regulatory elements involved in cancer progression, helping to link epigenetic changes with clinical outcomes. Moreover, the role of noncoding RNAs, including microRNAs and long noncoding RNAs (lncRNAs), has gained significant attention in epigenomics [104]. These molecules regulate gene expression and chromatin structure, which significantly impact tumor biology [104-109]. Moreover, RNA-seq has furthered the profiling of noncoding RNAs in various cancers, emphasizing their potential as biomarkers or therapeutic targets [110].

Bioinformatics tools are essential for analyzing the large datasets generated by epigenomic research. Bismark is often used for DNA methylation analysis, aligning sequencing data and identifying differential methylation regions at a single-base level [111]. MethylKit and DSS are popular for detecting methylation changes under different conditions in differential analysis [112, 113]. For ChIP-seq and ATAC-seq data, MACS is a favored tool for detecting peaks in chromatin accessibility, while SICER is particularly effective at revealing larger enrichment areas in histone modification studies [114, 115]. Tools like DiffReps and ChromHMM provide comprehensive analyses of chromatin states and accessibility across genomic regions, emphasizing the epigenetic patterns critical for understanding cancer biology [116, 117]. To separate meaningful biological signals from background noise, researchers frequently employ limma and EdgeR, R-based packages that assist in identifying significant epigenetic changes between cancerous and normal tissues [118, 119].

The integration of epigenomic data with bioinformatics creates numerous opportunities for identifying biomarkers in precision oncology. Epigenetic profiling has already revealed several promising biomarkers, including specific methylation patterns linked to certain cancer types and clinical outcomes [120]. For example, the methylation status of genes such as RASSF1A and GSTP1 has been investigated as a potential biomarker for diagnosing prostate cancer [121]. Additionally, epigenomic biomarkers can help in making therapeutic choices [122]. Genes exhibiting abnormal methylation can support the selection of patients for epigenetic treatments, while biomarkers related to drug response and resistance can inform personalized treatment plans based on an individual's epigenetic profile.

2.5 Metabolomics

Cancer is frequently characterized by altered metabolic pathways, particularly exemplified by the Warburg effect, wherein tumor cells predominantly favor glycolysis followed by lactic acid fermentation for energy production, even in the presence of oxygen [123]. This modification in metabolic processes facilitates enhanced cellular proliferation and contributes to tumor progression and metastasis. Metabolomics, defined as the comprehensive analysis of metabolites within biological systems, yields significant insights into the biochemical processes driving cancer biology [124]. Given that metabolites represent the end products of cellular activities, they reflect the physiological conditions of cells and tissues, thereby revealing critical information regarding metabolic dysregulation in cancer. Within the field of precision oncology, where treatments are tailored to individual patient profiles, metabolomics proves indispensable for identifying biomarkers [125]. Advancements in metabolomics and bioinformatics have already discovered several promising cancer biomarkers applicable for detection and prognosis. For example, elevated levels of metabolites such as 2-hydroxyglutarate have been correlated with specific cancers, including glioma and acute myeloid leukemia, thereby presenting new opportunities for early detection and personalized treatment [126, 127]. Furthermore, metabolomic profiling can enhance therapeutic strategies by identifying metabolites associated with drug resistance or sensitivity [128]. For instance, alterations in lipid metabolism have been linked to chemotherapy responses in breast cancer, suggesting that modifying treatments based on metabolic profiles may enhance efficacy [129].

Metabolomics has wide-ranging clinical applications in oncology. Metabolites can serve as early diagnostic markers, enabling the identification of cancer even in asymptomatic stages [130]. Additionally, metabolic profiles provide prognostic information, assisting in identifying patients at an elevated risk of recurrence or treatment resistance. Furthermore, metabolites associated with therapeutic response can inform personalized treatment strategies, enhancing targeted therapies’ effectiveness [128]. MS and nuclear magnetic resonance (NMR) spectroscopy are the principal techniques utilized for metabolomic profiling [131, 132]. MS is extensively used due to its sensitivity and capability to analyze complex mixtures. When integrated with chromatographic techniques such as gas chromatography (GC–MS) or liquid chromatography (LC–MS), MS facilitates both qualitative and quantitative analysis of a wide array of metabolites [133, 134]. This versatility makes MS an invaluable tool for investigating tumor metabolism. NMR, while less sensitive, offers distinct advantages, including nondestructive testing and the ability to ascertain metabolite structures. It is particularly beneficial for profiling metabolites within intact biological matrices, such as tissues or biofluids, and can corroborate findings from MS-based investigations due to its reproducibility and quantitative attributes.

The rapid expansion of metabolomics has led to vast datasets that require advanced computational and bioinformatics tools for effective analysis and interpretation. For instance, XCMS is commonly used for processing MS data, including peak detection, retention time adjustment, and metabolite quantification to uncover cancer-specific metabolic alterations [135]. Other popular tools such as MZmine and OpenMS facilitate the analysis of LC–MS and GC–MS liquid data, along with Sangerbox 3, which supports multiomics integration [136]. Furthermore, tools like MetaboAnalyst and MZmine play a crucial role in managing raw metabolomic data, performing peak identification, retention time adjustment, and normalization, thus ensuring high-quality data for further analyses [137, 138].

In metabolite identification, MetFrag is a valuable tool for in silico fragmentation and matching MS/MS data [139]. On the other hand, Global Natural Products Social (GNPS) provides web-based resources tailored for sharing and analyzing MS/MS data, particularly focused on natural products research [140]. Moreover, databases like the Kyoto Encyclopedia of Genes and Genomes (KEGG), the Human Metabolome Database (HMDB), and Reactome provide crucial information about biological pathways related to metabolites, aiding in pathway enrichment analysis [141, 142]. KEGG organizes pathways by merging genomic, chemical, and functional data, in cancer biology. Additionally, specialized software tools such as LipidSearch, LIMSA, and LipidBlast significantly improve the identification and quantification of lipids, thereby expanding the focus of metabolomic studies to include lipidomics, a vital component of cancer metabolism [143].

In metabolomics, statistical tools like R packages limma and glmnet, alongside software such as SPSS and SAS, help researchers conduct multivariate analyses, including principal component analysis (PCA) and partial least squares discriminant analysis [144, 145]. These methods aid in distinguishing metabolic patterns among various cancer types or clinical subgroups, revealing potential biomarkers linked to particular tumor traits or treatment responses.

Researchers are enhancing the identification of biomarkers that can substantially improve cancer diagnosis, prognosis, and therapeutic monitoring by integrating advanced bioinformatics tools with metabolomic analyses. Consequently, metabolomics presents significant potential for the future of precision oncology, paving the way for personalized treatment strategies (Table 1).

3.1 Data Preprocessing and Quality Control Tools

Data preprocessing and quality control are essential steps in analyzing high-throughput sequencing data, enhancing the integrity and reliability of subsequent analyses. The process typically includes several steps, such as trimming low-quality reads, removing sequencing adapters, and filtering out contaminants. This initial step is crucial for improving the integrity and reliability of subsequent analyses [168].

The quality control of sequencing data performed during the preprocessing phase. FastQC is a widely utilized tool for assessing the quality of raw sequencing data, generating comprehensive reports on base quality scores, GC content, and sequencing depth. Researchers utilize these reports to identify potential data issues, such as low-quality reads or biases, which may adversely affect the analysis [169].

Upon assessing the quality, Trimmomatic is used to trim low-quality bases and eliminate sequencing adapters, thereby ensuring the retention of only high-quality reads for subsequent analyses [170]. This procedure significantly enhanced the accuracy of variant calling and gene expression quantification, establishing a robust foundation for data interpretation. Following preprocessing, normalization is often essential to address variations attributable to discrepancies in sequencing depth or batch effects. ComBat and ComBat-ref are frequently employed to adjust for batch effects arising from technical variations in sample preparation or sequencing, thereby preventing false biomarker identification [171]. Furthermore, surrogate variable analysis (SVA) serves to detect and remove batch effects and other unwanted variations that are not directly measurable [172]. SVA is widely applied in RNA-seq and microarray studies to uphold data comparability across samples and experiments, ultimately improving the reliability of analytical results [173].

3.2 Biomarker Discovery Algorithms

Once the quality of data is assured, the subsequent step involves identifying potential biomarkers through computational algorithms that differentiate between relevant features (such as genes or proteins) and background noise. Feature selection methods, including Lasso and Elastic Net, are employed to mitigate the complexities of large datasets and to select the most predictive biomarkers [174]. Lasso is particularly advantageous for high-dimensional genomic data, while Elastic Net is adept at handling correlated datasets with interdependent variables [175, 176]. ML methods play a crucial role in biomarker discovery by uncovering intricate relationships within extensive datasets [177]. Random Forests amalgamate decision trees to facilitate robust feature selection, whereas support vector machines (SVMs) are utilized to differentiate between cancerous and noncancerous samples [178, 179]. Furthermore, deep learning techniques, especially in the context of multiomics and image data, unveil complex patterns through the application of neural networks [180].

The primary instruments used in ML include Random Forest and scikit-learn, with a focus on model training in R, along with TensorFlow and PyTorch for deep learning applications [181-185]. Furthermore, additional tools such as Weka, Bioinformatics Toolbox, and R/Bioconductor offer methodologies for statistical analysis [186, 187]. In addition, Cytoscape and GenePattern facilitate the visualization of biomarker interactions within pathways, thereby assisting in the interpretation of biomarker networks in the context of precision oncology [188, 189].

3.3 Multiomics Integration Platforms

Biomarkers are rarely identified through the utilization of a singular data type, as cancer encompasses intricate interactions across molecular strata. Multiomics integration platforms connect genomic, transcriptomic, proteomic, epigenomic, and metabolomic data, offering a holistic perspective on cancer biology and robust biomarkers supported by cross-omics evidence. Tools to integrate the multiomics data include SNF and iCluster. SNF constructs similarity networks for each omics category, which are then merged into a consensus network [190, 191]. This process facilitates biomarker discovery with consistent evidence across data layers. In comparison, iCluster groups multiomics data to identify cancer subtypes through common molecular traits, which may reveal biomarkers specific to subtypes that are relevant for prognosis or treatment response [192, 193].

Network-based methods are essential for identifying biomarkers. Tools like netboxr combine protein–protein interaction networks with genomic information, pinpointing regulatory pathways and biomarkers across various omics layers [194]. Cytoscape acts as a flexible platform to visualize molecular interaction networks, integrate multiomics data, and investigate biomarkers within biological pathways and network frameworks [188]. Moreover, Cytoscape serves as a platform for visualizing these networks and performing pathway enrichment analyses through various plugins, such as ClueGO and Bingo These tools enhance the biomarker discovery domain and encourage personalized strategies in oncology (Figure 2).

3.4 Pathway and Network Analysis

In cancer biomarker discovery, there is a growing emphasis on analyzing pathways and networks rather than solely examining individual genes or proteins. This transition recognizes the connections among molecular processes. Disruptions in biological pathways can occur due to altered gene expression or protein activity, and recognizing these changes may facilitate the development of robust biomarkers.

A range of tools enhances pathway enrichment and network development. Gene set enrichment analysis (GSEA) detects enriched gene sets or pathways crucial for understanding differential gene expression in biological contexts [195]. The database for annotation, visualization, and integrated discovery (DAVID) aids in functional annotation, highlighting enriched pathways and cellular components [196, 197]. The Reactome provides a curated collection of biological pathways and mapping tools that relate data to pathway disruption for biomarker identification [198, 199]. Ingenuity pathway analysis (IPA) is utilized for pathway analysis and network construction, widely applied in therapeutic biomarker research [200]. Network analysis helps identify biomarkers by examining genes or proteins within networks, focusing on key “hubs” that regulate vital processes. STRING creates protein–protein interaction networks and identifies interactions specific to cancer [155]. Another essential bioinformatics tool known as BioGRID used to display essential protein and genetic interactions for networks in cancer biology [201]. ClusterProfiler, an R package, enhances these tools by visualizing pathway enrichment and aiding in dataset interpretation [202].

Recently developed tools have significantly advanced the discovery of biomarkers using network-based approaches. For instance, NetPath meticulously curates the signaling pathways associated with cancer, facilitating the identification of biomarkers through the dysregulation of these pathways [203]. Metascape, on the other hand, consolidates various databases for pathway enrichment and network analysis, while GeneMANIA predicts gene functionality based on coexpression and interaction networks [204, 205]. Collectively, these tools empower researchers to identify biomarkers by systematically analyzing critical network nodes and dysregulated pathways.

3.5 Validation of Biomarkers

Validating biomarkers is a complex yet crucial process for translating research findings into clinical trials and applications [206]. Biomarker candidates that have successfully completed the initial verification phase and developed precise, robust quantitative assays must then undergo in silico, analytical, and clinical validation to ensure reliability and clinical relevance. In silico validation, which encompasses computational evaluations, has emerged as an integral part of the validation process for potential biomarkers [207]. This method employs mathematical modeling, simulations, and data analysis from publicly accessible datasets such as TCGA and Gene Expression Omnibus (GEO). These resources offer comprehensive multiomics and gene expression data from thousands of patients across a variety of cancer types. Specifically, TCGA plays a pivotal role in validating genetic and RNA-based biomarkers, while GEO emphasizes RNA-based biomarkers in a range of clinical conditions and cancers [208, 209]. Supplementary databases like ArrayExpress and the International Cancer Genome Consortium (ICGC) provide functional genomics validation through diverse experimental designs, allowing for the assessment of somatic mutations CNVs [210, 211]. Additionally, meta-analytic strategies enhance the validation process by combining data from numerous studies, thereby increasing statistical power and confirming the robustness of potential biomarkers [212, 213].

Bioinformatics tools offer numerous benefits, such as cost efficiency, comprehensive evaluations of potential biomarkers, and improved survival analyses alongside risk assessments in the validation process. Survival analyses, including Kaplan–Meier (KM) curve analyses utilizing TCGA and GEO datasets, allow researchers to link biomarker expression with prognostic results, including overall survival, progression-free survival, and disease-free intervals [214]. SurvExpress, an extensive gene expression database and online biomarker validation resource, enhances risk stratification by combining survival analysis with gene expression profiles, confirming the predictive value of biomarkers in separate cohorts [215]. For sophisticated statistical modeling, R-based packages like Survminer utilized to generate KM curves and Cox proportional hazards (Coxph) models [216]. The Coxph model is vital for examining the relationships between survival time and their predictors [217]. Web platforms such as GEPIA connect TCGA and GTEx data for gene expression and survival examination, investigating the prognostic significance of particular genes [218]. Additionally, tools like TIMER extend biomarker validation by factoring in immune cell infiltration dynamics with survival information, aiding in the assessment of how immune cells and gene expression affect survival rates [219]. Additionally, specialized bioinformatics platforms like OncoLnc and Sangerbox 3 streamline survival analyses utilizing TCGA-derived molecular profiles (e.g., mRNA, miRNA, lncRNA) to investigate survival outcomes [220]. Prognoscan offers meta-analysis capabilities to assess the connections between survival and gene expression, while the TCGA Biolinks (R package) facilitates large-scale survival research by maintaining reproducible access and analysis of TCGA data, thereby preserving methodological rigor in biomarker validation [221].

One essential aspect of bioinformatics tools involves performing biomarker analyses after an in vivo or in vitro efficacy study, which assists in validation efforts by correlating targets with the drug's mechanism of action [222]. This process helps predict the response or nonresponse of potential biomarkers. The findings from these analyses can serve as a basis for establishing inclusion and exclusion criteria for future clinical studies and validations. Bioinformatics tools systematically analyze downstream data from extensive in vitro assays, incorporating cell lines and genomically annotated tumor organoids, enabling researchers to differentiate between responding and nonresponding cell populations based on unique genomic signatures, thus providing early indications of possible biomarkers. By merging pharmacological response data from these screenings with computational models, researchers can link drug sensitivity or resistance with various genomic features such as gene expression profiles, mutational landscapes, CNVs, and pathway activation states, to formulate robust, multidimensional biomarker hypotheses. Promising candidates identified in vitro can progress to in vivo validation in preclinical models, where functional and mechanistic studies sharpen their biological significance [222]. Bioinformatics-based approaches play a crucial role in choosing xenograft or patient-derived xenograft models that align with a drug's molecular target or mutation profile, thereby ensuring translational relevance. This iterative process links in silico predictions to experimental validation, assisting in the prioritization and confirmation of the most clinically significant biomarkers potential [223].

Advanced analytical validation models, including AI and ML, enhance the validation process through cross-validation, which rigorously assesses the application of biomarkers across various datasets [224]. This enables evidence-based decisions about their clinical usefulness. Such an approach prevents overfitting, where a model too closely aligns with training data, impairing its ability to generalize to new data. Effective analytical validation assures both accuracy and generality of the model. Clinical validation is a crucial aspect of biomarker validation, as it improves the reliability of results and verifies the clinical significance of potential biomarkers by assessing their sensitivity and specificity [225]. Additionally, clinical validation clarifies the relationship between a biomarker and clinical aspects like treatment response, disease stage, and comorbid conditions. This validation relies on both in silico and analytical methods, which may involve retrospective reviews of past clinical trial data or new prospective trials. Retrospective reviews serve as a form of external clinical validation, particularly when biomarker assessments were not included in the original study design. In contrast, prospective clinical trials showcase the clinical relevance of a biomarker, acting as a kind of external validation that illustrates how its application can improve health results [226]. Various prospective clinical trial designs aim to confirm the clinical utility of biomarkers. A significant case is the United States Food and Drug Administration (US FDA's) 2017 tissue-agnostic approval of pembrolizumab, the first treatment authorized based on a biomarker rather than tumor location [227]. This decision stemmed from the KEYNOTE-016 study, which indicated higher overall response rates in patients with microsatellite instability-high (MSI-H) tumors treated with pembrolizumab compared with those with microsatellite stable (MSS) tumors, regardless of cancer type. Regulatory approval was based on pooled results from five trials (total N = 149), with MSI-H status retrospectively identified in 14 patients from two prospective studies, while 135 patients from three additional trials were prospectively confirmed. The objective response rate for MSI-H patients was 39.6% (including a 7% complete response rate) across 15 tumor types, which is deemed clinically significant. In contrast, patients with MSS tumors in the KEYNOTE-016 trial had a 0% response rate, underscoring the predictive power of the biomarker [228]. Another significant prospective clinical trial is the EURTAC trial, which resulted in the US FDA approval of erlotinib as a first-line therapy for metastatic NSCLC with EGFR mutations [229]. Other examples of such trials include the MARVEL trial and SWOG S0819 [230, 231].

4.1 Overview of AI Techniques in Oncology

AI techniques in biomedical research are categorized into supervised and unsupervised learning methods, both effective for revealing patterns in intricate datasets [256-258]. Supervised learning involves training algorithms using labeled data with known outcomes, which helps models understand the links between inputs (like gene expression profiles) and outputs (such as cancer subtypes). Common approaches include decision trees, random forests, SVMs, and neural networks, which are employed in cancer classification and in forecasting patient outcomes based on biomarker information. In contrast, unsupervised learning tackles unlabeled data to discover hidden patterns or clusters. Techniques like clustering (e.g., k-means and hierarchical clustering) and dimensionality reduction (e.g., PCA and t-SNE) are instrumental in biomarker discovery, identifying patient subgroups, and simplifying data complexity [181].

Deep learning, a branch of ML, plays a crucial role in biomarker discovery by seamlessly extracting hierarchical features from raw data [259]. CNNs and RNNs excel in the analysis of genomic, transcriptomic, and imaging data. These models have proven effective for classifying tumors based on histopathological images and predicting gene mutations from sequencing data, providing powerful tools for biomarker identification and large-scale exploration of biological patterns [260, 261].

4.2 AI Tools for Predictive Biomarkers

AI-powered tools for biomarker discovery and predictive modeling have expedited oncology research [262]. Predictive modeling utilizes ML techniques to examine the connections between molecular traits and clinical results, including survival rates and therapeutic responses. A prime example is DeepVariant, a deep learning tool created by Google, which identifies genetic variants from NGS data and enhances mutation detection in cancer genomics [263]. PandOmics, a cloud-based software platform that applies AI and bioinformatics approaches to multimodal omics data for biomarker discovery [264]. These tools are capable of predicting prospective biomarkers from a range of omics data (Figure 3).

Numerous specialized platforms have highlighted the critical role of AI in identifying biomarkers. OncoKB, a precision oncology knowledge base, merges genomic sequencing with clinical trial data to offer insights into gene mutations vital for cancer treatment [164]. Open-source tools like DeepLearning4J and Omics Data Science offer deep learning libraries for complex biomarker analysis [265]. Drug Discovery AI predicts effective drug combinations based on tumor biomarkers, thereby promoting progress in personalized oncology treatments.

4.3 Challenges and Opportunities

AI technologies are significantly impacting biomarker discovery in oncology, yet challenges remain [266]. A key obstacle is the scarcity of high-quality labeled data, particularly for rare cancers or newly identified biomarkers. Training effective AI models with small datasets can lead to overfitting, reducing both generalizability and clinical relevance. Although strategies like data augmentation, transfer learning, and cross-validation can help address these problems, the lack of data continues to restrict AI's potential in biomarker discovery. Another significant challenge involves integrating various data types, such as genomics, transcriptomics, proteomics, and clinical records, which present different formats and noise levels, thereby complicating model training and efficiency [267]. Advancements in data integration techniques are vital to boost AI's capability in identifying clinically meaningful biomarkers.

Despite facing challenges, AI demonstrates significant promise in precision oncology. It facilitates the integration of diverse datasets, uncovering complex biomarker signatures that consider molecular interactions. This could lead to more accurate biomarker identification, reducing the time and costs associated with discovering predictive markers, and enabling the application of findings in clinical practice. However, the “black box” characteristic of AI models raises interpretability concerns, making clinicians reluctant to adopt these technologies [268]. Exploring explainable AI and visualization techniques is essential for nurturing clinical trust. Additionally, leveraging patient data for AI applications introduces ethical and regulatory dilemmas, necessitating well-defined guidelines to uphold ethical integrity [269]. Collaborative efforts among researchers, clinicians, and data scientists can help tackle existing challenges, while advancements in AI technologies, like natural language processing and advanced imaging, offer fresh possibilities for biomarker discovery and therapeutic target identification [270].

5.1 Cancer Genomic Databases

Cancer genomic databases such as TCGA, ICGC, COSMIC, Genomic Data Commons (GDC), and cBioPortal are essential for examining genetic mutations (CNVs), and variations in gene expression associated with cancer [271]. TCGA is notable for its vast repository, which includes data from thousands of patients with various tumor types, featuring both genomic sequences and transcriptomic profiles [272]. ICGC focuses on diverse populations, aiding in the identification of genetic alterations relevant to specific demographics [273]. COSMIC specializes in somatic mutations, offering insights into actionable mutations and predictive capabilities [274]. Platforms like GDC and cBioPortal facilitate integrative analyses by combining data from numerous cancer studies and enhancing the connections between genetic alterations and clinical outcomes [274].

Additional resources will significantly enhance biomarker research beyond just genome-level changes. The Cancer Mutation Census gathers information on mutations in genes related to cancer, facilitating the identification of common mutations [275]. Although not cancer-specific, ExAC offers data on rare variants, contributing to a better understanding of cancer susceptibility [276]. The GEO contains extensive high-throughput gene expression data, including essential cancer studies that are vital for comparing expression patterns in cancerous and normal tissues [277]. Together, these resources create a strong foundation for biomarker discovery, encouraging further investigation into molecular variations in cancer.

5.2 Transcriptomic Databases

Transcriptomic databases are essential for providing insights into gene expression, allowing researchers to probe gene regulation and identify DEGs that serve as biomarkers. GEO is a key public repository that offers high-throughput gene expression data from various genomic techniques, such as microarray and RNA-seq. It supports investigations into differential expression related to different cancer types, treatment responses, and disease stages. Similarly, ArrayExpress provides curated datasets from microarray and sequencing studies, enabling researchers to explore transcriptional changes linked to specific cancer phenotypes [278]. These transcriptomic resources are vital for precision oncology, assisting identification of critical biomarkers for diagnosis, prognosis, and treatment responses [278].

5.3 Proteomics and Metabolomics Databases

Proteomics and metabolomics databases provide useful insights into protein expression, PTMs, and metabolite profiles, facilitating the identification of biomarkers at both protein and metabolite levels. The PRoteomics IDEntifications database serves as a vital resource for MS-based proteomics, offering essential data on protein expression, identification, and modifications that are crucial for discovering oncology biomarkers [279]. Furthermore, PeptideAtlas compiles high-confidence protein and peptide identifications, enhancing biomarker discovery by providing information on protein abundance, modifications, and cancer-related interactions [280]. Other essential proteomic databases and repositories include GPMDB, MassIVE, PASSEL, SRMAtlas, and Panorama utilizes for several studies including biomarker discovery [281]. MetaboLights offers small-molecule profiles from various biological samples, enabling researchers to explore metabolic changes in cancer and identify metabolite biomarkers reflective of tumor metabolism and its influence on cancer progression [282].

5.4 Integrated Multiomics Databases

Integrated multiomics databases improve the simultaneous analysis of genomic, transcriptomic, proteomic, and metabolomic data, providing a thorough understanding of cancer biology. These platforms facilitate the integration of various data types, helping researchers identify biomarkers across different molecular layers. cBioPortal, a key open-access platform, merges genomic, transcriptomic, and clinical data from large cancer studies like TCGA and ICGC, allowing for the investigation of mutations, expression patterns, and patient outcomes to uncover biomarkers. Another essential tool, Xenabrowser, offers interactive access to multiomics data from TCGA and ICGC, combining genomic, epigenomic, and transcriptomic data to identify clinically relevant biomarkers [23, 283].

6.1 Examples of Successful Biomarker Discovery

Numerous key biomarkers discovered through bioinformatics have transformed precision oncology. These biomarkers enhance cancer diagnosis and prognosis and guide therapies by identifying patients likely to respond to specific treatments. One significant example is PD-L1 (programmed death-ligand 1), which allows tumor cells to evade immune responses by attaching to the PD-1 receptor [300]. Bioinformatics methods, including differential gene expression analysis and pathway enrichment, have played a vital role in identifying PD-L1 as an immune checkpoint biomarker [301]. Examining high-throughput RNA-seq data with tools like DESeq2 and EdgeR have shown PD-L1 upregulation in various cancers [21]. By merging transcriptomic data with clinical outcomes and immune-related pathways through tools such as GSEA and IPA, researchers have gained a deeper understanding of PD-L1's role in immune evasion [302]. PD-L1 testing is instrumental in selecting patients who could benefit from immune checkpoint inhibitors like pembrolizumab and nivolumab, with bioinformatics crucial in associating PD-L1 expression with responses to immunotherapy [300].

A significant breakthrough in biomarker research is the discovery of BRCA1 and BRCA2 germline mutations, which strongly indicate hereditary breast and ovarian cancers [303]. Breast and ovarian cancers are responsible for the majority of the cancer-related deaths among women around the globe [304-306]. Bioinformatics analyses of extensive sequencing data have validated these mutations as highly penetrant biomarkers. Mutation detection tools, such as Mutect and GATK, were utilized to analyze whole-exome sequencing (WES) and NGS data to identify BRCA mutations [307]. Additionally, variant annotation tools like ANNOVAR and SnpEff, combined with computational predictions, have assessed the functional implications of these mutations [308]. The clinical impact is significant as BRCA testing now serves a vital role in evaluating cancer risk and shaping treatment decisions, especially with PARP inhibitors like olaparib, which target BRCA-mutated cancers [309]. Genetic testing for BRCA mutations has become the standard of care in oncology, particularly for patients diagnosed with breast and ovarian cancer [310].

6.2 Therapeutic Implications

The identification of predictive biomarkers significantly enhances our understanding of cancer biology and guides treatment decisions. These biomarkers enable the selection of targeted therapies, thereby facilitating customized treatments personalized to patients’ molecular profiles to attain optimal effectiveness.

EGFR mutations are associated with increased sensitivity to EGFR inhibitors such as erlotinib and gefitinib, especially in NSCLC [311]. Leveraging bioinformatic tools like VarScan and OncoKB is essential for identifying these mutations as actionable biomarkers [312]. By analyzing EGFR mutations via sequencing data, links to downstream pathways have been revealed, which assist in selecting targeted therapies and results in better clinical outcomes [313].

Rearrangements of anaplastic lymphoma kinase (ALK) in lung cancer have spurred the development of ALK inhibitors, such as crizotinib and alectinib [314, 315]. Bioinformatics tools, including FusionCatcher and STAR-Fusion, evaluate RNA-seq data for ALK fusions and network-based tools like Cytoscape assist in validating ALK partners [316]. Testing for ALK rearrangements now determines eligibility for ALK inhibitors, thereby enhancing progression-free survival rates in lung cancer [317].

TFF3 and pBAD are emerging as important targets in several malignancies including mammary, lung, liver, pancreatic, colorectal, ovarian, and endometrial due to their roles in tumor growth and treatment responses [318-337]. They are vital for epithelial cell regeneration and are linked to poor patient outcomes. By targeting TFF3, it may be possible to hinder tumor development by influencing the tumor microenvironment [338]. Tools such as GEPIA and DAVID are useful for assessing TFF3 expression and its related pathways. Moreover, pBAD, recognized for its ability to promote apoptosis, could be harnessed to increase cell death in tumors, making them more susceptible to chemotherapy [327]. Analyzing TFF3 and pBAD expression through bioinformatics reveals their contributions to cancer progression and supports innovative treatment strategies.

7.1 Data Integration and Heterogeneity

One of the significant challenges in biomarker discovery driven by bioinformatics is integrating various omics data types—such as genomics, transcriptomics, proteomics, epigenomics, and metabolomics—into a unified framework [340]. Each omics layer offers unique insights while differing in structure, scale, and complexity; for example, genomic data are typically static, whereas transcriptomic and proteomic data are more dynamic. Advanced computational tools are required to tackle issues such as discrepancies in data types, noise, and missing values to effectively integrate these datasets. Despite continuous efforts using tools like iCluster and SNF, integrating multiomics data remain a significant challenge due to a lack of standardization, variations in sample collection methods, and disparate data processing pipelines [192, 340]. Additionally, the heterogeneity of cancer complicates biomarker discovery even further, as differences among patients and tumor subtypes imply that signatures may not be applicable across the board [262].

7.2 Interpreting Big Data

The massive amount of data produced by high-throughput technologies poses a considerable challenge [341]. Current sequencing technologies generate terabytes of data, requiring advanced computational infrastructure for processing, storage, and analysis. Many bioinformatics tools demand significant computational power, which can be a barrier for researchers without access to high-performance systems [342]. The necessity for real-time analysis in clinical settings further increases the complexity. Proper long-term storage of omics data, particularly in large cohort studies, requires robust data management systems, raising issues around data retrieval, sharing, and reproducibility. Additionally, the increase in data volume raises the likelihood of false positives and misleading correlations, especially in omics studies with multiple variables are simultaneously tested [343]. Addressing these challenges requires enhanced interdisciplinary collaboration and more effective validation and regulatory processes to facilitate the integration of bioinformatics-driven biomarkers into clinical practice.

7.3 Translating Findings to Clinical Practice

Bioinformatics has significant potential for the discovery of biomarkers; however, applying these findings in clinical settings remains challenging [339]. Many biomarkers identified through bioinformatics do not transition to clinical practice due to issues with validation, regulatory hurdles, and the necessity to align computational findings with patient care. Comprehensive experimental validation, typically requiring large and well-documented patient cohorts, is crucial for bioinformatics insights, yet such groups are frequently scarce [344]. Additionally, obtaining US FDA regulatory approval is a lengthy and expensive endeavor that involves multiple clinical trials. The gap between bioinformatics tools used for discovery and those available in clinical laboratories adds further challenges, as computational methods need to be reproducible, transparent, and pertinent for clinical application. This situation is exacerbated by insufficient collaboration between bioinformaticians and clinicians, hindering the effective prioritization of clinically significant discoveries. To overcome these obstacles, enhanced interdisciplinary collaboration and more efficient validation and regulatory processes are essential for successfully integrating bioinformatics-derived biomarkers into clinical practice.

7.4 Clinical Integration of Omics Data in Precision Oncology

The integration of single-cell and spatial omics data alongside clinical information represents a significant advancement in personalized cancer therapy [33]. This approach correlates molecular profiles with clinical outcomes, facilitating the discovery of new biomarkers and significantly enhancing the prediction of treatment responses. By utilizing various data sources, the researchers can generate actionable insights that directly inform targeted treatment strategies, ultimately resulting to more effective and customized patient interventions.

To fully utilize the potential of emerging technologies in precision oncology, it is crucial to pursue ongoing advancements in bioinformatics tools and analytical frameworks. Future research should focus on key areas such as integrating multiomics data, improving visualization techniques, and applying ML and AI [262]. Specifically, integrating multiomics data require creating advanced computational methods that combine information from genomics, transcriptomics, proteomics, and metabolomics, thus building a comprehensive model of cellular function and disease progression. This integrated approach deepens our understanding of cancer biology and helps identify clinically relevant biomarkers. Moreover, enhancing visualization capabilities is vital for interpreting high-dimensional and complex datasets, making them more understandable and accessible for researchers and clinicians. By advancing visualization and computational methods, bioinformatics tools can yield deeper biological insights, improving patient stratification, diagnostic precision, and therapeutic decision-making. Tackling these bioinformatics challenges will enhance our grasp of cancer mechanisms and accelerate the translation of multiomics findings into practical clinical applications, thereby driving the progress of personalized healthcare [23].

Researchers in precision medicine emphasize the necessity of standardized frameworks and forward-looking clinical studies for ensuring reproducibility, quality assurance, and clinical relevance in bioinformatics methods [345]. By combining various data types, such as genomics, imaging, proteomics, and electronic health records (EHRs), there is a considerable opportunity to create and validate AI-based medical models that improve predictive diagnostics, disease categorization, and treatment strategies [346]. These standardized frameworks enhance reproducibility by setting validated protocols for bioinformatics processes and AI models, while ensuring adherence to regulations and reliability in evaluations. Furthermore, incorporating diverse data strengthens the generalizability of AI, mitigates bias, and facilitates the transfer of computational insights into practical clinical applications. However, challenges remain, including the intricacy and heterogeneity of multimodal data, the requirement for comprehensive, well-annotated clinical datasets, and the significant computational power needed for integration and model training [347]. Ethical and privacy issues regarding patient data also demand secure methods of sharing, like federated learning, to safeguard sensitive information while promoting collaborative research [269]. Additionally, regulatory challenges pose significant barriers to the validation and application of AI-powered clinical decision-support tools, making it necessary for collaborative efforts to standardize approaches across institutions and regulatory agencies [348].

Despite the existence of these challenges, addressing them through interdisciplinary collaboration among clinicians, bioinformaticians, data scientists, and regulatory experts is necessary to develop robust, interpretable, and clinically useful AI models [349]. While hurdles continue to persist, the advantages of precision medicine grounded in bioinformatics, such as enhanced patient stratification, biomarker identification, and personalized treatment options, are substantial. The ongoing innovation and advancement of these methodologies will continue to drive progress in cancer diagnostics and treatment strategies, ultimately transforming the future of personalized oncology care.

7.5 Ethical and Privacy Considerations in the Practical Application of Bioinformatics Tools in Precision Oncology

The integration of bioinformatics with AI and ML algorithms in clinical practice raises several vital ethical and privacy concerns [350]. Addressing these issues is crucial for protecting patient welfare, ensuring equitable access to care, and promoting the responsible use of advanced technologies. Ethical challenges in bioinformatics include data privacy, the importance of informed consent, data sharing, potential misuse of genetic data, the necessity for transparency in result interpretation, and algorithmic bias in AI and ML systems [351]. Establishing these ethical frameworks is critical for the effective use of biomarkers and designing personalized medicine while maintaining ethical standards.

Advanced sequencing and bioinformatics technologies produce massive amounts of genomic and clinical information, which risks confidentiality breaches if not properly safeguarded [352]. To preserve patient privacy and secure sensitive information, it is essential to anonymize or de-identify data before sharing or publication [353]. However, the unique nature of genomic data raises concerns about achieving complete anonymization. This challenge requires the implementation of robust protocols and advanced computational tools to thwart reidentification attempts. Data encryption is another vital measure, ensuring safe transmission and storage of patient information [354]. Additionally, encryption tools help guard against unauthorized access by staff or external parties. To enhance these efforts, enforcing strict access controls, using secure servers, following regulatory frameworks like GDPR (General Data Protection Regulation) and HIPAA (Health Insurance Portability and Accountability Act), and allowing only authorized personnel access to sensitive information can significantly mitigate the risk of privacy violations [355].

Informed consent is a key ethical issue in biomedical research [352]. Patients whose data are used in such studies must receive detailed information regarding the study's purpose, the use of their biological samples and personal data, and any associated risks and benefits [356]. In bioinformatics, obtaining meaningful consent requires addressing the complexities of information sharing, broad consent, and the need for reconsent [357]. Effectively communicating complex genomic data and analysis to participants can be challenging. Thus, researchers should create explicit, accessible consent materials free of jargon, ensuring that participants fully understand the study's objectives and methodologies. Some researchers may opt for broad consent or move to dynamic consent, which supports long-term scientific development and enables the future use of participants’ data in new research [358]. Furthermore, this approach can enhance scientific exploration, while necessitating a strong ethical commitment to respect participants’ autonomy, including their right to withdraw consent at any moment during the study.

Alongside privacy concerns, sharing data and ensuring equitable access to precision oncology studies and analyses are crucial [359]. Data sharing is vital in bioinformatics but requires solid agreements to maintain ethical standards and protect participants' rights [360]. Repositories such as dbGap (database of genotypes and phenotypes) offer dual access: open-access data for transparency and a controlled-access system for participant privacy protection [361]. Controlled access allows researchers to responsibly use sensitive data while ensuring participant confidentiality, thus aligning the benefits of shared data with ethical responsibilities. Precision oncology must ensure that innovations based on biomarkers are available to all patient demographics, including those from low socioeconomic or underrepresented backgrounds [362]. Social justice should guide funding and resource allocation decisions to prevent healthcare disparities. Additionally, predictive models, including AI and ML algorithms, play a vital role in bioinformatics, yet they also raise significant ethical issues that require careful consideration [363]. A key problem with AI systems is the biases inherent in training data, which can involve missing information, unrecognized patients (algorithmic bias), and sample size challenges and misclassification. For example, vulnerable populations, such as those with low socioeconomic status, psychosocial obstacles, and immigrants, often experience nonrandom data omissions in healthcare systems, leading to gaps in EHRs (e.g., absence of diagnostic tests, chronic illness medications, or social factors like housing instability) [351, 363]. Such deficiencies can lead ML algorithms to misinterpret existing data or exclude at-risk individuals from clinical tools designed for early intervention. Furthermore, EHRS might miss data on specific elements critical for enhancing health outcomes in these populations. Tackling these biases necessitates employing fairness-aware algorithms, utilizing diverse training datasets, and intentionally integrating metadata to prevent discriminatory treatment outcomes that could worsen socioeconomic healthcare disparities or disproportionately affect certain groups [363, 364]. Key steps include identifying or defining the target population, choosing training/testing datasets that embody this diversity, developing and validating algorithms across varied healthcare contexts, rigorously examining potential discriminatory patterns during data processing, establishing feedback mechanisms for validating output, and prioritizing clinically meaningful results instead of just performance benchmarks. These strategies are intended to promote equitable healthcare delivery and ensure that ML-based algorithms and tools effectively serve all populations, especially those typically marginalized in data representation [360].

The ethical aspects of bioinformatics are complex, influenced by technical details and societal duties. A key concern is the lack of clarity in AI models, which can hide decision-making processes. It is crucial to ensure transparency and accountability in AI-driven decisions to uphold trust, particularly given bioinformatics’ impact on research outcomes. Additionally, ethical considerations require careful attention to essential elements of ethical bioinformatics practices, including data privacy, informed consent, and responsible data sharing, alongside measures to prevent the misuse of participants’ genetic information [352, 360].