Abstract 690: Optimization of biopsy scheduling in clinical studies of T cell bispecifics through an integrated modeling and simulation, digital pathology and machine learning approach

• Published in: Cancer research (Chicago, Ill.) Vol. 79; no. 13_Supplement; p. 690
• Main Authors: Hutchinson, Lucy G., Soubret, Antoine, Ribba, Benjamin, Charoin, Jean-Eric, Phipps, Alex, Peck, Richard, Grimm, Oliver
• Format: Journal Article
• Language: English
• Published: 01.07.2019
• ISSN: 0008-5472; 1538-7445
• DOI: 10.1158/1538-7445.AM2019-690

Abstract In the context of cancer immunotherapy clinical trials, baseline and on-treatment tumor biopsies may provide important insight into whether a treatment is working as expected, and furthermore whether efficacy is anticipated. For tumor-retained antibodies that perturb the behaviour of immune cells, such as T cell bispecific antibodies (TCBs), spatial information derived from biopsy images may be particularly insightful. On-treatment biopsies in clinical trials are usually scheduled at a time point that is considered convenient for the study design and when therapeutic effects, such as T cell infiltration, are expected to be distinguishable in tumor tissue. To our knowledge, there have been no attempts to investigate optimal scheduling of on-treatment biopsy sample collection using quantitative approaches due to lack of clinical data at a sufficiently diverse range of time points. Our integrated tissue pathology, disease modeling and machine learning workflow is designed to select the time point at which on-treatment biopsies could be most informative for making reliable predictions of response to treatment. Leveraging around 20 baseline and on-treatment digitized biopsy images from patients undergoing treatment with immune stimulating TCBs, we train a mathematical model to simulate tumor cell/T cell interactions in the tumor microenvironment. The mathematical model produces an enriched dataset of “virtual” biopsy images corresponding to predictions at intermediate time points. The virtual biopsies are evaluated based on their ability to predict treatment response. Specific mechanisms of action of bispecific antibodies, such as upregulation of T cell activation and/or proliferation, are taken into account in the structure of the mathematical model. The model is tuned and validated using machine learning techniques, and a reserved “test” dataset comprising images that were not used to estimate model parameters is used to evaluate model performance. Our workflow has the potential to inform clinical study design by promoting a scientific basis for the selection of an on-treatment biopsy schedule. Future applications of this workflow include identification of tissue properties that may contribute to inter-individual variability, and simulations of novel doses and schedules for combinations of immune-modulating cancer therapies. Citation Format: Lucy G. Hutchinson, Antoine Soubret, Benjamin Ribba, Jean-Eric Charoin, Alex Phipps, Richard Peck, Oliver Grimm. Optimization of biopsy scheduling in clinical studies of T cell bispecifics through an integrated modeling and simulation, digital pathology and machine learning approach [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 690.