---

The following post is a consensus opinion from the following attendees of the 2025 Siracusa conference:

• Morten Anderson, PhD Roskilde University, Denmark
• Alexander R. A. Anderson, PhD Moffitt Cancer Centre, Florida, US
• David Basanta Gutierrez, PhD Moffitt Cancer Centre, Florida, US
• Angela Bentivegna, PhD University Milano-Bicocca, Italy
• Sebastien Benzekry, PhD Inria and Center for Research on Cancer of Marseille, France
• Sergio Branciamore, PhD City of Hope, California, US
• Sarah Brüningk, PhD University of Bern, Switzerland
• Martina Conte, PhD Politecnico University of Turin, Italy
• Farnoush Farahpour, PhD University of Duisburg-Essen, Germany
• Aleksandra Karolak, PhD Moffitt Cancer Centre, Florida, US
• Alvaro Köhn-Luque, PhD University of Oslo, Norway
• Tommaso Lorenzi, PhD Politecnico University of Turin, Italy
• Guillermo Lorenzo, PhD University of A Coruña, Spain
• Babgen Manookian, PhD City of Hope, California, US
• Russell Rockne, PhD City of Hope, California, US
• Andrei S Rodin, PhD City of Hope, California, US
• Lara Schmalenstroer, University of Duisburg-Essen, Germany
• Juan Soler, PhD University of Granada, Spain
• Cristian Tomasetti, PhD City of Hope, California, US
• Konstancja Urbaniak, PhD City of Hope, California, US

May 2025, the Palazzo Vermexio in Siracusa, Sicily, hosted "The Future of Cancer Research: The Interplay Between Machine Learning and Computational Modeling." This conference, initiated and hosted by City of Hope investigators, brought together early-career researchers and established experts in Mathematical Oncology to discuss the current and future impact of artificial intelligence technologies on the field. The focus was on immersive learning, spirited debate, and collaborative visioning, moving away from a typical lecture-heavy format. The two-day meeting was structured to help engagement and dialogue, with each day beginning with structured presentations led by early-career researchers and focused on a specific topic. That was followed by group discussions. To give you an idea of what these presentations were about, here are some brief highlights:

• Molecular Dynamics and AI: Aleksandra Karolak opened the meeting with a session on how AI enhances molecular dynamics simulations, enabling more accurate predictions of protein-protein and drug-target interactions. This fusion of atomistic modeling with machine learning is accelerating drug discovery and offering new insights into resistance mechanisms at the molecular level.
• Bayesian Networks in Biomarker Discovery: Konstancja Urbaniak and Babgen Manookian then explored how static and dynamic Bayesian networks can model molecular interactions and transitions over time, especially when integrated with deep learning or molecular dynamics simulations.
• AI in Oncology: Sarah Brüningk presented on how machine learning is transforming oncology—from diagnosis to treatment optimization—by enabling precision medicine through deep learning and reinforcement learning models, with a focus on applications to medical imaging modalities such as MRI.
• Dynamic Mathematical Modeling: Martina Conte and Lara Schmalenstroer discussed how mathematical models, including differential equations and agent-based simulations, can predict tumor invasion and treatment response.

Discussion Themes

With the help of these presentations to guide brainstorming, these are some of the themes that emerged:

1. AI and the Future of Scientific Discovery

A central (and for some of us, worrying) question was whether AI could eventually outpace human researchers in generating new knowledge. Several scenarios included for example: Could AI design viable molecules that don’t exist in nature? Could it discover new biological or physical laws or mathematical conjectures? While some of us viewed this as speculative, others among us pointed to real-world examples where AI has already begun to uncover novel insights from complex datasets where the ability of the human mind to comprehend structures in the data is limited.

Yet, many among us noted that AI’s capabilities are bounded by the data it learns from. In domains where data is sparse or non-existent, human intuition and creativity remain irreplaceable. This led to a broader reflection: perhaps the future role of researchers is not to compete with AI, but to use AI as an accelerating tool to work at the frontier of knowledge and science—where data is scarce, uncertainty is high, new paradigms are needed, and where AI will be inherently limited. Indeed, the central goal of science is about identifying and asking the next question, a task that currently AI is not well equipped to address.

2. Agentic AI and Multi-Scale Modeling

Agentic AI, or the use of multiple LLMs as ‘agents’ to execute multi-component workflows, was identified as an advance that necessitates a paradigm shift in how researchers in the field of Mathematical Oncology operate. Our group discussed the inevitable development of workflows that mimic the process of mathematical modeling: take experimental or clinical data, construct a family of plausible mechanistic mathematical models, simulate top candidate models, and generate computational codes to reproduce the results in a matter of seconds. Such agentic workflows could be a transformative change to the way Mathematical Oncology research is conducted.

The importance of integrating black box approaches such as AI with mechanistic understanding, especially at the molecular level was raised as a potential opportunity. Resistance mutations, protein flexibility, and drug-target interactions are often overlooked in high-level models. By embedding structural and biochemical data into AI systems, researchers can build multi-scale models that connect atomic-level changes to tissue-level dynamics and patient outcomes. For instance, the concept of temperature is an emergent property that is not defined for an individual molecule. This multi-scale abstraction not only enhances predictive accuracy but also improves interpretability—an essential feature in clinical decision-making. The challenge, as several participants noted, is educational: how do we train the next generation to think across scales and disciplines?

3. Temporal Dynamics and Data Limitations

Several contributors highlighted the limitations of current machine learning models in capturing temporal dynamics. While ML excels at pattern recognition, it often assumes static or stationary data—an unrealistic assumption in oncology, where tumors evolve and patient responses change over time. ML and DL approaches that do not assume stationarity tend to be computationally demanding for real-world data. Even for methods that do handle time-series data, ML and DL approaches rely on prior observations and are therefore inherently limited in extrapolating to new situations often encountered in scientific and clinical research.

Mechanistic models, by contrast, are inherently dynamic and hypothesis driven. One promising direction was the use of mechanistic simulations to generate synthetic data for training deep reinforcement learning models. This hybrid strategy could reduce dependence on large clinical datasets while preserving mechanisms and hypotheses to be studied.

4. Ethics, Uncertainty, and Abstraction

Critical ethical and philosophical questions were also discussed. As AI systems increasingly rely on large datasets, for example from national health registries, and begin to be applied to highly temporally resolved patient-specific wearable data, how do we ensure patient autonomy and data sovereignty? What frameworks are needed to govern the use of sensitive health data in model training?

The group also wrestled with the concept of probability in medicine. Does probability even exist for single events such as tumor progression? For example, if tumor progression is the predicted outcome with 90% probability, is the prediction wrong if the tumor does not progress? Is probability a real property of biological systems, or merely a tool for managing uncertainty? This led to a broader discussion on abstraction that led all the way to the existence of free will. The discussion was brought back to practical reality by noting that while simplification is necessary for modeling, excessive abstraction can obscure critical biological complexity and ignore the reality that probabilities are a necessary means of describing model predictions. The consensus was clear—models must strike a balance between tractability and fidelity.

5. Education and the Next Generation

A recurring theme was the need to cultivate "multi-scale thinkers"—scientists who can navigate from molecular biology to systems-level modeling, and from theoretical abstraction to clinical application. The importance of mentorship, interdisciplinary training, critical thinking, and logical reasoning, for the next generation of scientists was strongly emphasized. Teaching students that the central goal of science is about asking the next question is an essential ingredient for career success in science, that for now remains a uniquely human endeavor.

There was also concern about the misuse of AI tools by students—and faculty—who lack foundational understanding. As AI becomes more accessible, the risk of over-reliance grows. The group called for a renewed focus on education that emphasizes conceptual depth, ethical awareness, and scientific rigor.

Looking Ahead

This 2025 conference in Siracusa was not a regular academic meeting—it was a microcosm of the current and potential future state of Mathematical Oncology research. It showcased the power of interdisciplinary collaboration, the promise of AI-augmented discovery, and the value of human insight.

P.S. This blog post was drafted by Microsoft Copilot based on the meeting agenda, notes, and a series of email exchanges between attendees following the meeting. The content was human generated and the text was edited and enhanced by the authors. Given the disclosure of the LLM use and human origin of the content and final product, we implore you to ask yourself, is the use of AI a problem here? Why or why not? Respond in the comments section and help shape the conversation about use of AI and LLMs in our field.