Matteo Pedone, research fellow at the Department of Statistics, Computer Science, Applications “G. Parenti” of the University of Florence, was our guest at the D2 Seminar Series. He presented his talk on “A Bayesian nonparametric approach to personalized treatment selection“.
Precision medicine is an approach to disease treatment that defines treatment strategies based on the individual characteristics of the patients. Motivated by an open problem in cancer genomics, in his work Matteo Pedone develop a novel model that flexibly clusters patients with similar predictive characteristics and similar treatment responses; this approach identifies, via predictive inference, which one among a set of therapeutic strategies is better suited for a new patient. The proposed method is fully model-based, avoiding uncertainty underestimation attained when treatment assignment is performed by adopting heuristic clustering procedures, and belongs to the class of product partition models with covariates, here extended to include the cohesion induced by the normalized generalized gamma process. The method performs particularly well in scenarios characterized by large heterogeneity among the predictive covariates in simulation studies. A cancer genomics case study illustrates the potential benefits in terms of treatment response yielded by the proposed approach. Finally, being model-based, the approach allows estimating clusters’ specific random effects and then identifying patients that are more likely to benefit from personalized treatment.
Would you explain briefly what is a Bayesian nonparametric approach and what has to do with personalized medicine?
Bayesian statistics is an approach to statistical inference based on Bayes’ theorem, where available knowledge about parameters in a statistical model is updated with the information in observed data. Bayesian statistics represents a flexible and comprehensive learning paradigm for modeling relationships in data, taking into account prior knowledge and uncertainty.
Bayesian nonparametric (BNP) is a branch of Bayesian statistics that allows for the number of parameters in a statistical model to be flexible and adapt to the data rather than being fixed in advance. BNP models are best suited when the data have complex, multi-modal distributions or non-linear relationships between variables; and when the model needs to adapt and update as new data become available.
Personalized medicine is a medical approach that uses individual genetic, environmental, and lifestyle information to tailor diagnoses and treatments for better patient outcomes. BNP models can be used in precision medicine to model individual patient data and make personalized predictions, such as prognoses or treatment responses. Since BNP models can learn complex relationships in the data and account for inter-patient variability, this approach leads to more accurate predictions than traditional, fixed-parameter models.
What are the needs the method you propose in your study is meeting?
Individualized treatment rule (ITR) in precision medicine refers to the use of patient-specific data and genomic information to guide the treatment selection, tailored to the individual’s unique characteristics. ITRs are crucial because patients can have different genetic profiles or heterogeneous responses to the treatment, and a one-size-fits-all approach may not be effective for all patients. In cancer genomics, defining ITRs is particularly challenging because tumors are characterized by large heterogeneity.
We propose an ITR that leverages information from historical patients to select the optimal treatment for a new patient. Our method uses a model-based approach to cluster patients based on their predictive biomarkers and treatment responses, allowing for the identification of the most suitable therapeutic strategy for a new patient. The new patient is assigned to the treatment that ensured the largest benefit to the historical patients with whom the new patient shows the largest similarity in terms of predictive biomarkers. Our approach overcomes the limitations of existing procedures, leading to improved treatment response and personalized treatment recommendations.
What is the major challenge you encountered in addressing this research?
BNP -as the whole Bayesian approach- is fascinating, but it can be challenging from a mathematical point of view. Additionally, the computational algorithms used in Bayesian nonparametric are often complex. The main challenge I faced was to get such a theoretical paradigm working in an applied framework and dealing with “real” data.
How do you think the approach you propose can potentially impact disease treatment in the future?
I hope that our research can be a useful contribution to those interested in statistical methodology for precision medicine.
You can access the video recording of his seminar here (registration needed).