Harshini Suresha

Machine Learning Approaches for Cardiovascular Risk Prediction in Cohort Data

ML & Cardiovascular Risk Prediction

Year2025
LocationIISc
TagsDeepSurv, Random Survival Forests, TCR, single-cell

Interpretable risk prediction from longitudinal cohorts, combining clinical, immune and single-cell features in Random Survival Forests and DeepSurv, judged by C-index, Brier score, calibration and decision curves.

Jul 2025 – Sep 2025 at IISc, Bangalore, supervised by Payel Roy.

Workflow figureOpen full size ↗
Figure for Machine Learning Approaches for Cardiovascular Risk Prediction in Cohort Data
Methods
  1. 01
    Data

    Longitudinal cohort data combining clinical, lifestyle, laboratory and immunological features, plus single-cell transcriptomic profiles, TCR repertoires and APOB-specific immune responses.

    Cohort data, scRNA-seq, TCR repertoires
  2. 02
    Representation

    Dimensionality reduction and representation learning to capture high-dimensional immune states.

    Dimensionality reduction, Representation learning
  3. 03
    Models

    Classical and nonlinear models for risk, including time-to-event modelling.

    Logistic Regression, SVM, Random Forest, Gradient boosting, Neural networks, Random Survival Forests, DeepSurv
  4. 04
    Evaluation

    C-index, integrated Brier score, time-dependent AUC, calibration curves, decision-curve analysis, cross-cohort transfer and feature-set ablation.

    C-index, Brier score, Calibration, Decision curves
  5. 05
    Longitudinal alignment

    Aligned clinical and immune trajectories to handle irregular sampling and changing baselines over time.

    Time-series alignment
← Back