NCI Data Jamboree (Project Abstract Submission): Submission #5

Presenter Information
---------------------
First Name: OSCAR
Middle Initial: MARINO 
Last Name: Vidal 
Degree(s): Ph.D.
Position/Title/Career Status: PI
Organization: Universidad del NOrte 
Organization Address:
Barranquilla. Atlantico 

Email: oorjuela@uninorte.edu.co

Additional Authors
------------------
List of Additional Authors:
- First Name: Jorge 
  Last Name: Velez
  Affiliation: UNINORTE


Abstract Information
--------------------
Abstract Category: Evaluating data quality for reproducibility and AI-readiness
Abstract Keywords: Precision Oncology Machine Learning Multi-Modal Data Integration Predictive Modeling
Abstract Title: AI-Driven Patient Stratification Models for Personalized Treatment in Breast Cancer Using Multi-Omic and Clinical Data
Abstract:
Breast cancer is a highly heterogeneous disease characterized by substantial variability in molecular profiles, treatment response, recurrence risk, and survival outcomes. Current clinical stratification approaches based on receptor status and tumor staging do not fully capture this complexity, often leading to suboptimal treatment selection. This project aims to develop and evaluate artificial intelligence (AI)-driven patient stratification models that identify clinically meaningful breast cancer subgroups using multi-omic and clinical data. The primary scientific question is whether integrated machine learning approaches can improve prediction of treatment response and patient outcomes compared with conventional classification methods.

During the 3-day jamboree, the team will construct and compare unsupervised and supervised learning frameworks for patient stratification. Planned analyses include clustering of patients based on genomic, transcriptomic, and clinical features; identification of molecular signatures associated with treatment response; and development of predictive models for outcomes such as overall survival, disease-free survival, and therapeutic response. We will evaluate the interpretability of resulting models using explainable AI techniques to identify key biomarkers and pathways driving subgroup assignments.

Developing reproducible workflows for integrating heterogeneous biomedical datasets can benefit researchers, clinicians, and data scientists working on cancer prognosis, biomarker discovery, and personalized medicine. The resulting analytical framework could be adapted to other malignancies and disease areas.Data Types and Repositories

publicly available datasets, including:

Clinical data: patient demographics, tumor characteristics, treatment records, and survival outcomes from the The Cancer Genome Atlas Breast Invasive Carcinoma (BRCA) cohort.
Transcriptomic data: RNA-sequencing gene expression profiles from The Cancer Genome Atlas.
Genomic data: somatic mutation and copy number variation data from The Cancer Genome Atlas.
Proteomic data (optional): protein expression profiles from Clinical Proteomic Tumor Analysis Consortium.
Validation datasets: independent breast cancer cohorts from Gene Expression Omnibus and/or International Cancer Genome Consortium.