Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Indication Finding: a novel use case for representation learning

Maren Eckhoff, Valmir Selimi, Alexander Aranovitch, Ian Lyons, Emily Briggs, Jennifer Hou, Alex Devereson, Matej Macak, David Champagne, Chris Anagnostopoulos

TL;DR

This work uses representation learning to generate embeddings of indications and prioritize them based on their proximity to the indications with the strongest available evidence for the MoA, demonstrating the successful deployment of this approach for anti-IL-17A.

Abstract

Many therapies are effective in treating multiple diseases. We present an approach that leverages methods developed in natural language processing and real-world data to prioritize potential, new indications for a mechanism of action (MoA). We specifically use representation learning to generate embeddings of indications and prioritize them based on their proximity to the indications with the strongest available evidence for the MoA. We demonstrate the successful deployment of our approach for anti-IL-17A using embeddings generated with SPPMI and present an evaluation framework to determine the quality of indication finding results and the derived embeddings.

Indication Finding: a novel use case for representation learning

TL;DR

This work uses representation learning to generate embeddings of indications and prioritize them based on their proximity to the indications with the strongest available evidence for the MoA, demonstrating the successful deployment of this approach for anti-IL-17A.

Abstract

Many therapies are effective in treating multiple diseases. We present an approach that leverages methods developed in natural language processing and real-world data to prioritize potential, new indications for a mechanism of action (MoA). We specifically use representation learning to generate embeddings of indications and prioritize them based on their proximity to the indications with the strongest available evidence for the MoA. We demonstrate the successful deployment of our approach for anti-IL-17A using embeddings generated with SPPMI and present an evaluation framework to determine the quality of indication finding results and the derived embeddings.

Paper Structure

This paper contains 16 sections, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Results
  3. Introduction to the approach
  4. Evaluation Framework
  5. Evaluation of Indication Finding performance
  6. Quantitative evaluation of ranked list
  7. Qualitative evaluation of ranked list
  8. Evaluating the quality of embeddings for the anti-IL-17A MoA
  9. Performance in prediction tasks
  10. Interpretability of embeddings: comparison to ICD-10 categories
  11. Discussion
  12. Methods
  13. Set up of approach
  14. Feature definition
  15. Embedding approach
  16. ...and 1 more sections

Figures (4)

  • Figure 1: Overview of the indication finding approach.
  • Figure 2: Recall of positive validations and negative validations for the top 50, top 100, and top 200 ranked indications. There were five positive and five negative validations, respectively.
  • Figure 3: Diagnosis embeddings mapped to two dimensions using UMAP and color-coded by the ICD-10 category to which they belong. Left: Plot shows all diagnosis features. Region 1: Alongside the grouping of hepatobiliary diseases, from diseases of the digestive system, portal vein thrombosis and oesophageal varices, from the diseases of the circulatory system category map closely. These circulatory diseases are often diagnosed as complications of hepatobiliary diseases, such as chronic liver diseases. Right: Plot filtered to show only diagnosis features from diseases of the digestive system ICD-10 category. Region 2: demonstrates a subcategory of diagnoses relating to oral and salivary gland disorders, e.g., stomatitis, oral cysts, diseases of the tongue. Region 3: demonstrates a subcategory of diagnoses relating to intestine disorders, e.g., intestinal malabsorption, vascular disorders of the intestine.
  • Figure 4: Waterfall to illustrate the impact of inclusion / exclusion criteria on cohort size.