# BioCOMPASS: Integrating Biomarkers into Transformer-Based Immunotherapy Response Prediction

Source: [BioCOMPASS: Integrating Biomarkers into Transformer-Based Immunotherapy Response Prediction](<https://www.sota2.com/research/paper/69cdf8490636e5bf4afccd7d>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-03T05:00:04.568Z

Authors: Sayed Hashim, Frank Soboczenski, Paul Cairns

Published: 2026-04-01T11:06:29.000Z

## 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.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Classification on Immunotherapy cohorts 8 (test)](<https://www.sota2.com/research/sota/classification-on-immunotherapy-cohorts-8-test.md>): Classification; Immunotherapy cohorts 8 (test); Accuracy: 70.

- [Classification on Immunotherapy cohorts Small (\<50 samples) 4 cohorts](<https://www.sota2.com/research/sota/classification-on-immunotherapy-cohorts-small-50-samples-4-cohorts.md>): Classification; Immunotherapy cohorts Small (\<50 samples) 4 cohorts; Accuracy: 69.77.

- [Classification on Immunotherapy cohorts Large (\>50 samples) 4](<https://www.sota2.com/research/sota/classification-on-immunotherapy-cohorts-large-50-samples-4.md>): Classification; Immunotherapy cohorts Large (\>50 samples) 4; Accuracy: 70.24.

- [Immunotherapy Response Prediction on LOCTO (Leave-one-cancer-type-out)](<https://www.sota2.com/research/sota/immunotherapy-response-prediction-on-locto-leave-one-cancer-type-out.md>): Immunotherapy response prediction; LOCTO (Leave-one-cancer-type-out across BLCA, KIRC, SKCM, STAD); Accuracy: 70.05.

- [Immunotherapy response prediction on LOTO across PD-1, PD-L1, CTLA-4, CTLA-4 + PD-1](<https://www.sota2.com/research/sota/immunotherapy-response-prediction-on-loto-across-pd-1-pd-l1-ctla-4-ctla-4-plus-pd-1.md>): Immunotherapy response prediction; LOTO across PD-1, PD-L1, CTLA-4, CTLA-4 + PD-1; Accuracy: 73.49.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.00739>)

[PDF](<https://arxiv.org/pdf/2604.00739.pdf>)
