P-NET Project
Dr. Maricel Kann, Department of Biological Sciences.
Tristan Godard, Department of Biological Sciences.
P-NET is a biologically informed deep learning model designed for cancer prediction that incorporates prior biological knowledge into its architecture. P-NET utilizes the Reactome database as the structural backbone of the neural network. A key limitation of Reactome is that it is built on canonical, well-studied biological pathways that are manually curated from existing literature, constraining the model to known Biology. Cancer biology is often driven by non-canonical interactions, which means tumor-specific interactions may not be present in those manually curated pathway. In contrast, STRING database does not rely on manually curated pathways but rather integrated experimentally validated protein-protein interactions, genomic context, and predicted functional associations. Since STRING also covers nearly the entire human proteome, it can capture a larger fraction of cancer-associated genes, such as those present in COSMIC. Replacing Reactome-based pathway organization with STRING-derived interaction clusters would improve gene coverage and better capture cancer-relevant, non-canonical protein interactions that are absent from curated pathways, while maintaining classification performance.
Artificial Intelligence for Precision Oncology: A Multi-Omics Knowledge Network for Personalized Lung Cancer Treatment
Dr. Maricel Kann, Department of Biological Sciences (2024).
Asma Toor, Department of Biological Sciences (2026).
Lung cancer, particularly non-small-cell lung cancer (NSCLC), is a significant health concern, with an estimated 238,340 new cases and 127,070 deaths expected in the United States in 2023 alone. Despite advances in targeted therapies and immune checkpoint blockade (ICB), resistance to these treatments is a common occurrence, underscoring the need for novel research approaches. This project aims to address these challenges by developing an artificial intelligence (AI) framework that integrates complex molecular data for the stratification of NSCLC patients. The proposed AI framework will enable personalized therapeutic strategies, overcoming the limitations of current precision oncology approaches. By translating knowledge from literature, functional genomics, and multi-omic patient-specific data, the framework will provide a comprehensive molecular model.
p53 Disease Variants Classification
Dr. Maricel Kann, Department of Biological Sciences
Sai Vallurupalli, Department of Biological Sciences (2020)
Functional classification of disease variants is crucial when utilizing precision medicine to improve personalized treatments of patients. Multiple studies have considered molecular links between genes. However, none of them include specific functional effects of gene variants. The purpose of this research is to manually classify variants in genes leading to diseases into gain- or loss-of-function. This will allow for an automatized prioritization of disease-causing genes using network analysis methods that model the gain and loss differently. In order to accomplish this, we will first identify a set of genes of interest that we would like to include in our study. We will then proceed to identify functional keywords that allows us to classify the type of change a variation results in. For instance, if we choose to classify the variants in the oncogene TP53, we will identify the types of variations as either conformational changes or DNA binding changes. Finally, we will classify the specific variants reported in different loci of the genes based on the identified changes in function. This will allow for differential treatments of disease variants with distinct effects, thereby improving prognosis and treatment of diseases.