BACKGROUND: It is estimated that over 720,000 people are living with bladder cancer in the United States. Many of these patients could benefit from immune-based therapies, such as adoptive T cell therapy (ACT) that uses patients’ autologous tumor-infiltrating T lymphocytes (TILs). The first clinical feasibility trial with ACT-TIL in bladder cancer led by Dr. Michael Poch will open at Moffitt Cancer Center this year. Here, we present a machine learning-based protocol that will help to determine whether a patient will be a good candidate for this immunotherapy. METHODS: We developed a Machine Learning Predictor of the Expansion of Tumor Infiltrating Lymphocytes (ML-PETIL) that assesses whether or not the tumor isolated from a patient will grow T cells. ML-PETIL used 56 retrospectively collected patients’ data consisting of demographic, clinical, or biological tumor sample (BTS) features (103 in total). Dimensionality reduction of the feature space was performed using a suite of machine learning algorithms and an ensemble method was applied to identify the set of robust features. Using these robust features, ML-PETIL determined the algorithm with the best performance metrics from a collection of optimized classifiers that discriminated between the Grow TILs and No TILs classes. RESULTS: We compared the Area under the Receiver Operating Characteristic Curve (AUC) for four predictors that used 4 different datasets. Predictor 1 – used robust demographic features only, and we reported an AUC of 0.58. Predictor 2 – used robust demographic and clinical features and yielded an AUC of 0.5. Predictor 3 – used robust demographic, clinical, and BTS features, resulting in an AUC of 0.68. Predictor 4 – used robust tumor digest BTS features related to immune cell repertoire, and we reported an AUC of 0.5. Our results show that a data-informed approach like ML-PETIL can identify the optimal combination of demographic, clinical, and BTS features that are predictive of TIL expansion from patients’ bladder tumors. CONCLUSION: The protocol we propose in this work is an efficient way to combine data-informed machine learning algorithms with patients’ data routinely collected in clinic. This tool may assist clinicians in assessing patient stratification for ACT-TIL therapy.