Using Artificial Intelligence to Predict Interactions between Immune and Tumor Cells

ASPIRE Award (2019-2022)

Alexander Baras, MD, PhD, Johns Hopkins University

Alexander Baras, MD, PhD

Immune cells can detect cancer-causing genetic mutations when mutated peptides known as neoantigens are presented on the surface of tumor cells. However, tumor cells may evade immune cell detection by a variety of mechanisms involving the presentation and recognition of those neoantigens. Because the protein complexes responsible for displaying these peptides on the cell surface can vary widely from patient to patient, studying these mechanisms and developing prognostic and predictive models remains a challenge. In addition, there is often inconsistency among the different clinical assays currently used to identify and understand each cancer patient’s unique pattern of mutations. Researchers in the Baras lab are using the computational power of artificial intelligence to tackle the challenges arising from this complexity. First, they are creating machine-learning algorithms that reconcile differences between clinical test platforms to more accurately understand the characteristics of tumors. Then they will extend their analyses to identify which mutated peptides and protein complexes are most likely to appear on the surface of tumor cells as influenced by the unique genetic make-up of both the tumor and the patient. Once they have achieved this goal, they will use machine learning to predict the ability of immune cells to detect and kill an individual patient’s tumor cells. This artificial intelligence-based approach promises to be useful in accurately predicting prognosis and forecasting how patients will respond to anti-cancer therapies, especially those that harness the immune system.


Sidhom JW, Siddarthan IJ, Lai BS, Luo A, Hambley BC, Bynum J, Duffield AS, Streiff MB, Moliterno AR, Imus P, Gocke CB, Gondek LP, DeZern AE, Baras AS, Kickler T, Levis MJ, Shenderov E. Deep learning for diagnosis of acute promyelocytic leukemia via recognition of genomically imprinted morphologic features. NPJ Precis Oncol. 2021.

Sidhom JW, Larman HB, Pardoll DM, Baras ASDeepTCR is a deep learning framework for revealing sequence concepts within T-cell repertoiresNat Commun. 2021.

Sidhom JW, Oliveira G, Ross-MacDonald P, Wind-Rotolo M, Wu CJ, Pardoll DM, Baras AS. Deep learning reveals predictive sequence concepts within immune repertoires to immunotherapy. Sci Adv. 2022.

Anaya J, Sidhom JW, Cummings CA, Baras AS. Probabilistic mixture models improve calibration of panel-derived tumor mutational burden in the context of both tumor-normal and tumor-only sequencing. Cancer Res Commun. 2023.