Abstract
Cancer is not a single disease, it is a complex and heterogeneous disease which leads to the second cause of death worldwide. Although all cancers manifest themselves as an uncontrolled growth of abnormal cells, they are actually distinct neoplastic diseases that possess different genetic and epigenetic alterations, underlying molecular mechanisms, histopathologies and clinical outcomes. Understanding the origins and growth of cancer requires understanding the role of genetics in encoding proteins that form phenotypes and molecular alterations at multiple levels (e.g., gene, cell, and tissue).
Introduction
This editorial, published in Frontiers in Physiology, synthesizes a specialized research topic focused on the intersection of mathematics, computation, and oncology. As cancer remains a leading cause of global mortality, the scientific community is increasingly turning toward multiscale modeling, machine learning, and systems biology to decode its inherent complexity. The goal of these computational strategies is to move beyond "one-size-fits-all" protocols toward precise, patient-specific therapies that account for the unique genetic and epigenetic landscapes of individual tumors.
The Challenge of Multi-Scale Heterogeneity
The primary obstacle in modern oncology is the integration of massive, multi-level biological data—spanning from single genes and proteins to whole tissues and clinical outcomes. Tumors are not static; they are dynamic, evolving entities that change over time and space. This evolution often leads to therapeutic resistance, where a treatment that was initially effective becomes obsolete as the tumor adapts. To address this, researchers require mathematical frameworks capable of simulating these complex interactions to predict how a specific patient’s tumor will progress or respond to various drug combinations.
Methodological Innovations
The research reviewed in this editorial highlights a diverse array of computational methodologies designed to extract actionable insights from high-dimensional "omics" data:
- Machine Learning and Deep Learning: Advanced algorithms, including Convolutional Neural Networks (CNNs) and Natural Language Processing (NLP) tools like Facebook’s fastText, are being repurposed to identify protein families (such as SNARE proteins) directly from amino acid sequences.
- Network Analysis: Techniques like Fuzzy Weighted Recurrence Networks (FWRN) and Weighted Gene Co-expression Network Analysis (WGCNA) are utilized to identify "hub genes"—critical nodes in biological networks that serve as potential biomarkers for cancers such as Adrenocortical Carcinoma (ACC) and Renal Cell Carcinoma (ccRCC).
- Mathematical Modeling:
- ODEs and PDEs: Ordinary and Partial Differential Equations are employed to model the kinetics of oncolytic viral therapy and the transport of drugs across the blood-brain barrier.
- Agent-Based Models (ABM): These hybrid models simulate the spatio-temporal dynamics of individual cells within the tumor microenvironment, providing a granular view of tumor growth.
- Hopfield Networks: Recurrent neural networks are used to identify "attractors" in gene expression data that correlate with patient survival rates.
- Parameter Estimation: The introduction of Conditional Robust Calibration (CRC) offers a Bayesian approach to calibrating non-linear models in systems biology, providing higher accuracy than traditional methods like ABC-SMC.
Key Insights: Evolution and Resistance
Several studies within this topic challenge traditional paradigms of cancer progression:
- Induced Resistance: Research suggests that drug-resistant clones may not always be present prior to treatment. Instead, the treatment itself, interacting with the tumor microenvironment, may induce resistance in previously sensitive cells.
- Early Metastasis: Mathematical modeling of colorectal cancer (CRC) lung metastases indicates that cancer cells may begin spreading 4–5 years before the primary tumor is clinically diagnosed. This finding has profound implications for early detection and screening strategies.
- The Equivalent Dose Metric: To better measure drug efficacy, researchers proposed a biophysically-based "equivalent dose" metric. This metric incorporates pharmacokinetic and pharmacodynamic (PK/PD) properties, offering a more precise measurement than standard concentration levels.
- Phenotypic Plasticity: Investigations into the NFATc protein reveal its role as a "stability factor" that allows cancer cells to maintain a hybrid state between epithelial and mesenchymal types, a key driver of metastasis and therapy resistance.
Quantitative Results and Clinical Potential
The application of these models has yielded significant results:
- Prognostic Signatures: A 24-hub-gene signature was identified in ACC that significantly correlates with both Disease-Free Survival (DFS) and Overall Survival (OS).
- Predictive Scoring: In Triple-Negative Breast Cancer (TNBC), a deep-learning analysis of CD8+ T cell immunofluorescence images successfully derived a prognostic score for early relapse.
- Cancer Subtyping: A new method based on Consensus Factor Analysis (CFA) has outperformed existing techniques in discovering novel cancer subtypes with distinct survival profiles across 30 different cancer types.
Conclusion and Future Directions
While these computational advances represent a significant leap forward, several challenges remain. The "black box" nature of some deep learning models necessitates a push toward greater interpretability for clinical use. Furthermore, many of these models are currently in silico prototypes; the next phase of research must focus on validating these tools in prospective clinical trials. As models become increasingly complex, the development of user-friendly interfaces, such as the ABC-GWAS database for breast cancer genetic variants, will be essential for integrating these powerful tools into standard oncological practice.
Summary updated