Cognitive Computing Approach Based Analysis of Genomic Information in Blood Plasma for Diagnosis and Prognostic for Cancer

Context & Background

Liquid biopsy is a revolutionary non-invasive diagnostic method that analyzes cell-free DNA (cfDNA) and cell-free RNA (cfRNA) in blood plasma. Analyzing genomic patterns in cfDNA enables early-stage cancer detection and tracking of treatment response without surgical tissue biopsy.

Problems to be Addressed

Integrating multi-modal cfDNA signatures (methylation and fragmentation patterns) is computationally difficult. Models suffer from batch effects when training on data from different laboratories, leading to high false-positive rates.

Aims and Objectives

1. Develop sensitive computational methods to detect cancer from cfDNA/cfRNA profiles.
2. Create self-learning models using adaptive transfer learning to predict tissue of origin.
3. Validate findings using blood samples from Indian cancer patients.

Methodology

The project curates published cfDNA datasets and processes them using reference-free deconvolution methods. Blood and biopsy samples from cancer patients are collected at AIIMS. Library preparation and sequencing are performed at IIIT-Delhi. Adaptive transfer learning adjusts the models to local lab variables with minimal training data.

Expected Outcomes

Non-invasive liquid biopsy diagnostic tools, biomarkers for cancer tracking, training of PhD students, and partnerships with clinical diagnostic companies.