Biotherapy

Overview

This episode of Inside Cancer Careers from the National Cancer Institute highlights the pioneering work of computational scientists developing artificial intelligence (AI) tools to improve cancer treatment outcomes. Dr. Eytan Ruppin, Dr. Tiangen Chang, and Dr. Yingying Cao from NCI's Center for Cancer Research discuss their AI-based model, LORIS, designed to predict patient response to immune checkpoint blockade therapy, a form of cancer immunotherapy that has transformed cancer care but benefits only a subset of patients. The conversation delves into the model's development, clinical relevance, limitations, and future directions in AI-driven cancer research, as well as the scientists' personal career journeys and advice for those interested in computational biology and medicine.

What is Cancer Data Science?

Dr. Eytan Ruppin explains the role of cancer data scientists in 2024, emphasizing the exponential growth of cancer-related data across multiple dimensions. Computational analysis is essential to interpret this data to understand cancer biology and to develop new therapeutic strategies. His lab focuses on identifying key data features that can guide cancer therapy decisions, including predicting patient responses to various treatments. The goal is to assist clinicians in selecting the most effective therapies tailored to individual patients, complementing but not replacing clinical judgment.

LORIS: Predicting Immunotherapy Response with AI

Immunotherapy, particularly immune checkpoint blockade (ICB), has revolutionized cancer treatment, yet only about 20-30% of patients respond positively. Identifying which patients will benefit remains a critical challenge. To address this, Drs. Chang and Cao developed LORIS (LOgistic Regression-based Immunotherapy-response Score), a machine learning model that predicts patient response probability to ICB therapies using six clinical variables routinely collected in standard care.

• Model Design and Variables: LORIS is based on logistic regression, a statistical method that converts input variables into a probability score between 0 and 1. The six variables include tumor mutational burden (TMB), albumin levels, patient age, prior therapy history, and others selected for their availability and clinical relevance.
• Clinical Utility: The model is designed to be practical for clinicians, using data that are commonly measured, thus facilitating integration into clinical workflows.
• Validation and Accuracy: LORIS has been validated retrospectively on multiple patient cohorts, achieving approximately 75-80% accuracy in predicting response. However, the team stresses the importance of prospective clinical trials to confirm its predictive power without bias.
• Improvement Over Existing Biomarkers: Tumor mutational burden, an FDA-approved biomarker with a threshold of 10 mutations per megabase, is included in LORIS but recognized as suboptimal alone. The model identifies responders who would be missed by TMB alone, potentially expanding the pool of patients who could benefit from immunotherapy.
• Applicability Across Cancer Types: LORIS captures common predictive features across multiple cancers. A lung cancer–specific version incorporates PD-L1 expression, a known immunotherapy biomarker relevant to that cancer type.

Understanding Logistic Regression in LORIS

Logistic regression is a statistical technique that models the probability of a binary outcome—in this case, response or non-response to immunotherapy—based on input variables. Unlike linear regression, which predicts values across an infinite range, logistic regression constrains predictions between 0 and 1, representing probabilities. The LORIS model weighs each clinical variable to calculate a response probability score, aiding clinicians in treatment decisions.

Limitations and the Need for Prospective Validation

The LORIS model was developed and tested using retrospective data, which means the outcomes were already known during model training. This can introduce bias, as models may inadvertently be optimized to fit past data. Prospective studies, where the model is tested on new patients in real time without prior knowledge of outcomes, are necessary to rigorously evaluate LORIS's clinical utility and ensure unbiased performance.

Future Directions in AI and Cancer Research

The team discusses exciting ongoing and future research avenues that build on and extend beyond LORIS:

• Pathology Slide Analysis: Leveraging AI to analyze traditional hematoxylin and eosin (H&E) stained pathology slides to predict molecular tumor characteristics such as gene expression and mutations. This approach could democratize access to molecular data without the need for expensive sequencing technologies.
• Tumor Microenvironment and Blood Biomarkers: Integrating single-cell RNA sequencing and blood-based biomarkers to better understand the tumor immune microenvironment and improve prediction of immunotherapy response. For example, Dr. Chang's recent work identifies B-cell signatures in both tumor tissue and blood that correlate with response.
• Multi-Modal Data Integration: Combining clinical, molecular, spatial, and imaging data to build more comprehensive and accurate predictive models.
• Global Health Impact: Applying AI tools in resource-limited settings, such as parts of Africa, to improve cancer diagnosis and treatment accessibility. The team collaborates with international partners to deploy these technologies where deep sequencing is not widely available.

Personal Career Journeys and Advice

The guests share their diverse paths into computational cancer research, highlighting the interdisciplinary nature of the field:

• Dr. Eytan Ruppin: Originally trained in medicine and psychiatry, he transitioned to computer science inspired by the book "Gödel, Escher, Bach" by Douglas Hofstadter. After years in computational neuroscience, he shifted focus to cancer research, finding it a more tractable problem with impactful outcomes.
• Dr. Yingying Cao: Began in preventive medicine, then moved into bioinformatics during graduate studies. She is fascinated by treating the human body as a quantitative system and focuses on tumor microenvironment interactions using advanced molecular techniques.
• Dr. Tiangen Chang: Started in plant science and transitioned to cancer research motivated by personal family experiences with cancer. He values persistence, trust in oneself, and focusing on translationally relevant scientific questions.

They emphasize the importance of following scientific curiosity, choosing meaningful research questions, embracing interdisciplinary collaboration, and cultivating kindness and teamwork. Dr. Ruppin advises aspiring scientists to be brave, persistent, and to prioritize impactful research over publication quantity.

Recommendations and Resources

• Podcasts: "Intelligence Squared US" for engaging debates on science, technology, society, and culture, which can sharpen critical thinking.
• Books: "World Views, An Introduction to the History and Philosophy of Science" by Richard DeWitt, recommended for understanding the evolution of scientific thought and the provisional nature of scientific knowledge.
• Popular Science: New Scientist magazine, favored by Dr. Ruppin for accessible updates on physics and other sciences.
• Online Tool: The LORIS model is available online for clinicians to input patient data and obtain immunotherapy response predictions, facilitating practical application.
• NanCI App: An AI-powered app from NCI that helps researchers navigate scientific literature, discover relevant papers, and build professional networks.

Source Information

This content is based on the National Cancer Institute's Inside Cancer Careers podcast episode titled "AI and Immunotherapy: A Breakthrough in Cancer Treatment," released on July 18, 2024. The episode features in-depth interviews with Dr. Eytan Ruppin, Dr. Tiangen Chang, and Dr. Yingying Cao from NCI's Center for Cancer Research. The transcript provides detailed insights into the development and application of the LORIS AI tool, the future of AI in cancer research, and career advice for aspiring scientists. The discussion reflects the speakers' perspectives and ongoing research efforts without making unverified claims.