Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Personalised Drug Identifier for Cancer Treatment with Transformers using Auxiliary Information

Aishwarya Jayagopal, Hansheng Xue, Ziyang He, Robert J. Walsh, Krishna Kumar Hariprasannan, David Shao Peng Tan, Tuan Zea Tan, Jason J. Pitt, Anand D. Jeyasekharan, Vaibhav Rajan

TL;DR

The paper addresses the scarcity of patient-level drug response data by introducing PREDICT-AI, a transformer-based framework that predicts drug efficacy from sparse diagnostic-panel mutations while explicitly modeling the variable-length mutation sequences. It employs a novel two-stage training regime, TransformerMTLR for survival prediction and TransformerDRP for drug response, augmented by a two-tier tokenization that encodes gene and mutation-level features and by incorporating auxiliary survival information into the learning process. Empirical results show state-of-the-art performance on survival (CI improvements over baselines) and DRP benchmarks (AUROC $=64.96\%$, AUPRC $=84.85\%$) with notable drug-specific gains, along with ablation evidence that pretraining and survival supervision meaningfully boost accuracy. The authors also implement a treatment recommendation system deployed at a clinical site to assist MTBs, discuss deployment challenges, and outline lessons for trust-building and indirect evidence usage in clinical decision making. The work advances personalized oncology by integrating sequential genomic inputs, auxiliary outcomes, and real-world clinical deployment to guide targeted therapies.

Abstract

Cancer remains a global challenge due to its growing clinical and economic burden. Its uniquely personal manifestation, which makes treatment difficult, has fuelled the quest for personalized treatment strategies. Thus, genomic profiling is increasingly becoming part of clinical diagnostic panels. Effective use of such panels requires accurate drug response prediction (DRP) models, which are challenging to build due to limited labelled patient data. Previous methods to address this problem have used various forms of transfer learning. However, they do not explicitly model the variable length sequential structure of the list of mutations in such diagnostic panels. Further, they do not utilize auxiliary information (like patient survival) for model training. We address these limitations through a novel transformer based method, which surpasses the performance of state-of-the-art DRP models on benchmark data. We also present the design of a treatment recommendation system (TRS), which is currently deployed at the National University Hospital, Singapore and is being evaluated in a clinical trial.

Personalised Drug Identifier for Cancer Treatment with Transformers using Auxiliary Information

TL;DR

The paper addresses the scarcity of patient-level drug response data by introducing PREDICT-AI, a transformer-based framework that predicts drug efficacy from sparse diagnostic-panel mutations while explicitly modeling the variable-length mutation sequences. It employs a novel two-stage training regime, TransformerMTLR for survival prediction and TransformerDRP for drug response, augmented by a two-tier tokenization that encodes gene and mutation-level features and by incorporating auxiliary survival information into the learning process. Empirical results show state-of-the-art performance on survival (CI improvements over baselines) and DRP benchmarks (AUROC , AUPRC ) with notable drug-specific gains, along with ablation evidence that pretraining and survival supervision meaningfully boost accuracy. The authors also implement a treatment recommendation system deployed at a clinical site to assist MTBs, discuss deployment challenges, and outline lessons for trust-building and indirect evidence usage in clinical decision making. The work advances personalized oncology by integrating sequential genomic inputs, auxiliary outcomes, and real-world clinical deployment to guide targeted therapies.

Abstract

Cancer remains a global challenge due to its growing clinical and economic burden. Its uniquely personal manifestation, which makes treatment difficult, has fuelled the quest for personalized treatment strategies. Thus, genomic profiling is increasingly becoming part of clinical diagnostic panels. Effective use of such panels requires accurate drug response prediction (DRP) models, which are challenging to build due to limited labelled patient data. Previous methods to address this problem have used various forms of transfer learning. However, they do not explicitly model the variable length sequential structure of the list of mutations in such diagnostic panels. Further, they do not utilize auxiliary information (like patient survival) for model training. We address these limitations through a novel transformer based method, which surpasses the performance of state-of-the-art DRP models on benchmark data. We also present the design of a treatment recommendation system (TRS), which is currently deployed at the National University Hospital, Singapore and is being evaluated in a clinical trial.
Paper Structure (25 sections, 7 equations, 8 figures, 4 tables)

This paper contains 25 sections, 7 equations, 8 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Background
  4. Related Work
  5. Data
  6. Method
  7. Preliminaries
  8. Tokenization
  9. TransformerMTLR: Multi-Task Logistic Regression- based Survival Prediction
  10. TransformerDRP: Pre-trained Transformer-based Response Prediction
  11. Experiments and Results
  12. Survival Prediction with TransformerMTLR
  13. Comparison with state-of-the-art models
  14. Ablation Tests
  15. Deployment in a Clinical Setting
  16. ...and 10 more sections

Figures (8)

  • Figure 1: Overview of our personalized Treatment Recommendation System (TRS) and its role in clinical treatment planning.
  • Figure 2: Overview of the PREDICT-AI, consisting of three main components: (A) TransformerMTLR: a multi-task logistic regression model based on transformers for survival prediction. (B) TransformerDRP: a pretrained transformer-based drug response prediction model incorporating the AUDRC prediction task. (C) Detailed description of the transformer encoder layers.
  • Figure 3: Overview of proposed tokenization procedure with gene and mutation tokenizers.
  • Figure 4: Effects of individual components in the PREDICT-AI model.
  • Figure 5: Left panel shows mutations present in patient genomic profile. Middle and right panels displays top 10 recommendations with supporting evidence as boxplots.
  • ...and 3 more figures