BioCOMPASS: Integrating Biomarkers into Transformer-Based Immunotherapy Response Prediction

2026-04-01

AI Summary

Current models for predicting immunotherapy response exhibit limited generalizability across cohorts, cancer types, and treatment regimens, primarily due to the small scale and high heterogeneity of clinical data. Building upon the COMPASS framework, this work proposes a knowledge-integrated Transformer architecture that innovatively treats biomarkers as supervisory signals rather than input features. The model incorporates a treatment-gating mechanism and a pathway consistency loss, leveraging a customized loss function to align intermediate representations with established biological knowledge. Evaluated under rigorous cross-validation strategies—including leave-one-cohort-out, leave-one-cancer-type-out, and leave-one-treatment-out—the proposed approach demonstrates significantly improved generalization performance, thereby validating the efficacy of knowledge-guided representation learning in immunotherapy response prediction.

Abstract

Datasets used in immunotherapy response prediction are typically small in size, as well as diverse in cancer type, drug administered, and sequencer used. Models often drop in performance when tested on patient cohorts that are not included in the training process. Recent work has shown that transformer-based models along with self-supervised learning show better generalisation performance than threshold-based biomarkers, but is still suboptimal. We present BioCOMPASS, an extension of a transformer-based model called COMPASS, that integrates biomarkers and treatment information to further improve its generalisability. Instead of feeding biomarker data as input, we built loss components to align them with the model's intermediate representations. We found that components such as treatment gating and pathway consistency loss improved generalisability when evaluated with Leave-one-cohort-out, Leave-one-cancer-type-out and Leave-one-treatment-out strategies. Results show that building components that exploit biomarker and treatment information can help in generalisability of immunotherapy response prediction. Careful curation of additional components that leverage complementary clinical information and domain knowledge represents a promising direction for future research.

Problem

• immunotherapy response prediction
• generalisability
• biomarkers
• small datasets
• cross-cohort evaluation

Innovation

• transformer-based model
• biomarker integration
• self-supervised learning
• generalizability
• immunotherapy response prediction