Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Integration of Graph Neural Network and Neural-ODEs for Tumor Dynamic Prediction

Omid Bazgir, Zichen Wang, Ji Won Park, Marc Hafner, James Lu

TL;DR

This work proposes a heterogeneous graph encoder that utilizes a bipartite Graph Convolutional Neural network (GCN) combined with Neural Ordinary Differential Equations (Neural-ODEs) and is able to discover a tumor dynamic model that significantly improves upon an empirical model which is in current use.

Abstract

In anti-cancer drug development, a major scientific challenge is disentangling the complex relationships between high-dimensional genomics data from patient tumor samples, the corresponding tumor's organ of origin, the drug targets associated with given treatments and the resulting treatment response. Furthermore, to realize the aspirations of precision medicine in identifying and adjusting treatments for patients depending on the therapeutic response, there is a need for building tumor dynamic models that can integrate both longitudinal tumor size as well as multimodal, high-content data. In this work, we take a step towards enhancing personalized tumor dynamic predictions by proposing a heterogeneous graph encoder that utilizes a bipartite Graph Convolutional Neural network (GCN) combined with Neural Ordinary Differential Equations (Neural-ODEs). We applied the methodology to a large collection of patient-derived xenograft (PDX) data, spanning a wide variety of treatments (as well as their combinations) on tumors that originated from a number of different organs. We first show that the methodology is able to discover a tumor dynamic model that significantly improves upon an empirical model which is in current use. Additionally, we show that the graph encoder is able to effectively utilize multimodal data to enhance tumor predictions. Our findings indicate that the methodology holds significant promise and offers potential applications in pre-clinical settings.

Integration of Graph Neural Network and Neural-ODEs for Tumor Dynamic Prediction

TL;DR

This work proposes a heterogeneous graph encoder that utilizes a bipartite Graph Convolutional Neural network (GCN) combined with Neural Ordinary Differential Equations (Neural-ODEs) and is able to discover a tumor dynamic model that significantly improves upon an empirical model which is in current use.

Abstract

In anti-cancer drug development, a major scientific challenge is disentangling the complex relationships between high-dimensional genomics data from patient tumor samples, the corresponding tumor's organ of origin, the drug targets associated with given treatments and the resulting treatment response. Furthermore, to realize the aspirations of precision medicine in identifying and adjusting treatments for patients depending on the therapeutic response, there is a need for building tumor dynamic models that can integrate both longitudinal tumor size as well as multimodal, high-content data. In this work, we take a step towards enhancing personalized tumor dynamic predictions by proposing a heterogeneous graph encoder that utilizes a bipartite Graph Convolutional Neural network (GCN) combined with Neural Ordinary Differential Equations (Neural-ODEs). We applied the methodology to a large collection of patient-derived xenograft (PDX) data, spanning a wide variety of treatments (as well as their combinations) on tumors that originated from a number of different organs. We first show that the methodology is able to discover a tumor dynamic model that significantly improves upon an empirical model which is in current use. Additionally, we show that the graph encoder is able to effectively utilize multimodal data to enhance tumor predictions. Our findings indicate that the methodology holds significant promise and offers potential applications in pre-clinical settings.
Paper Structure (21 sections, 7 equations, 3 figures, 3 tables)

This paper contains 21 sections, 7 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Heterogeneous Graph Encoder
  4. Information Fusion
  5. Tumor Volume Prediction
  6. Tumor Volume Encoder
  7. Neural-ODE.
  8. Results
  9. Reproducing Tumor Volume Data
  10. Tumor Volume Dynamics Prediction
  11. Conclusion
  12. Patient-derived Xenograph (PDX) dataset
  13. Modified RECIST categories
  14. Comparing GCN with non-GCN models.
  15. Results
  16. ...and 6 more sections

Figures (3)

  • Figure 1: Model architecture overview.A) Integration of GCNs and Neural-ODEs for tumor dynamics prediction. B) Heterogeneous Graph Encoder: Two bipartite graph attention convolution NNs extract disease-gene and drug-gene associations and a graph convolution NN extracts gene-gene interactions. The output of all three are integrated into the embedding representing the baseline (pre-treatment) state.
  • Figure 2: Tumor dynamics prediction: (A) A comparison of tumor volume fitting between tumor growth inhibition (TGI) model proposed by zwep2021identification and our proposed NODE model. (B) In each of the 3 rows: tumor volume prediction using the proposed model for an individual PDX instance is shown. The blue curves represent the measurements (ground-truth), the green curves represent our proposed model predictions, and the red curves represent the TGI model predictions. Each of the column corresponds to a single choice of observation window (of increasing size from left to right): the highlighted blocks in dark blue blue indicate the observation windows beyond which our proposed model correctly captures the classification of the mRECIST response category as the ground-truth.
  • Figure A.1: Predictive performance of our proposed model as a classifier for mRECIST categories, with and without the heterogeneous graph encoder and considering different lengths of observation windows.