Details

Autor(en) / Beteiligte

• Firooz, Ali
• Funkhouser, Avery T.
• Martin, Julie C.
• Edenfield, W. Jeffery
• Valafar, Homayoun
• Blenda, Anna V.

Titel

Abstract 5425: Analysis of cancer patients’ molecular and clinical data using artificial intelligence and machine learning approaches

Ist Teil von

• Cancer research (Chicago, Ill.), 2023-04, Vol.83 (7_Supplement), p.5425-5425

Erscheinungsjahr

2023

Quelle

EZB Electronic Journals Library

Beschreibungen/Notizen

• Abstract Background: Development and clinical course of cancer is multifactorial with influences from the general health status of the patient, germline and neoplastic mutations, co-morbidities, and environment including lifestyle. For effective and individualized treatment of each patient, such multifactorial data must be easy-to-access and easy-to-analyze. Purpose Statement: Cancers are characterized on a molecular level by the presence of complex gene mutations and other specific molecular markers. Moreover, special importance is placed on so-called cancer-critical genes, mutations of which are involved in the development and progression of various cancers. However, not all detected sequence alterations in these genes are known as cancer-causing mutations; thus, a more detailed and sensitive analysis is prudent. In addition, there is a limited number of established and reliable cancer biomarkers of sera. To that end, a complete analysis of molecular basis of cancer needs to include additional biomarkers such as galectins and glycans, since patients’ galectin and glycomic profiles have promising cancer differentiating and diagnostic potential. Methods: We utilized a Relational Database Management System populated by clinical data from the Prisma Health Cancer Institute Biorepository of ~6,000 cancer patients with at least 66 different cancer diagnoses. Molecular data is available for gene mutations, serum galectin proteins, and glycomic profiles of cancer patients. Mutation status of 50 cancer-critical genes in 1,500 patients, 320 individual patient profiles of 5 serum galectin proteins, and serum and tissue glycomic profiles of 60 patients have been included and will be expanded. In addition, healthy control values for galectin and glycomic profiles were obtained and added for reference. We performed statistical and AI models of Data Analytics using R, Python, and TensorFlow platforms. A comprehensive set of patient data was used to develop a predictive model of patient outcome using the clinical observations as the desired outcome. Results: The use of typical statistical analyses (linear and logistic regression) revealed insignificant correlation between the predictors and the cancer type of patient outcome. However, the use of the Decision Tree revealed some interesting relationships that can be used for explainability and reliability of the Machine Learning approaches. Finally, Artificial Neural Network approaches provided the best performance in classification of cancer types from the given information. Conclusion: Our studies provide predictive models that could potentially be used to improve the diagnostic and prognostic power of data collected from patients at presentation. However, the dichotomy of black box AI approaches that perform better than explainable approaches, complicate deployment of these techniques in the domain of medicine and healthcare. Citation Format: Ali Firooz, Avery T. Funkhouser, Julie C. Martin, W. Jeffery Edenfield, Homayoun Valafar, Anna V. Blenda. Analysis of cancer patients’ molecular and clinical data using artificial intelligence and machine learning approaches. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5425.